Sepsis interrogation method and system based on multi-round dialogue

Abstract

This invention provides a sepsis consultation method and system based on multi-turn dialogue, belonging to the field of intelligent consultation equipment technology. This invention achieves effective integration of standardized medical knowledge and personalized clinical experience through the construction of a two-dimensional knowledge graph and cross-graph interactive retrieval, improving the accuracy and reliability of sepsis diagnosis and treatment recommendations. Through structured sepsis diagnosis and treatment chain constraints and a confidence-driven dynamic strategy, it alleviates the illusion problem of large language models in complex clinical reasoning, ensuring that the generated content conforms to clinical diagnosis and treatment logic norms. Through semantic anchor guidance, hierarchical activation, and knowledge feedback mechanisms, it achieves multi-granular semantic association from macro-clinical scenarios to micro-medical concepts, balancing retrieval accuracy and knowledge coverage. The confidence evaluation system comprehensively considers knowledge sufficiency, redundancy, and consistency, adaptively balancing knowledge utilization and information exploration, improving the reasoning efficiency and diagnosis and treatment quality of multi-turn dialogue.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical consultation equipment technology, specifically to a sepsis consultation method and system based on multi-turn dialogue. Background Technology

[0002] Sepsis is a systemic inflammatory response syndrome caused by infection and is one of the leading causes of death in patients in the intensive care unit (ICU). Early identification and timely intervention are crucial for improving the prognosis of sepsis patients. However, the clinical manifestations of sepsis are complex and diverse, involving dysfunction of multiple organ systems. Its diagnosis relies on the physician's comprehensive judgment of the patient's symptoms, signs, laboratory tests, and imaging data, which places extremely high demands on the professional skills of clinicians.

[0003] With the rapid development of electronic health records, clinical guidelines, and medical knowledge bases, AI-based medical question-answering systems are gradually becoming important tools for assisting clinical decision-making. Existing medical question-answering methods are mainly based on generative pre-trained language models, but when dealing with highly specialized sepsis diagnosis and treatment questions, they are prone to factual errors and logical inconsistencies, creating a kind of "illusion." Therefore, researchers have proposed a knowledge-enhancing generation strategy that guides language model generation by integrating structured medical knowledge graphs, thereby improving the credibility of answers to some extent.

[0004] However, existing methods still have significant limitations when dealing with complex diseases like sepsis, which involve multiple systems, multiple etiologies, and require multi-level reasoning: on the one hand, they lack fine-grained modeling of real clinical diagnosis and treatment scenarios, and the generated content is too generalized or deviates from the core diagnosis and treatment goals; on the other hand, the retrieval mechanism is unable to quickly extract targeted knowledge that conforms to individual patient differences and cannot fully capture the deep semantic structure of the problem. Summary of the Invention

[0005] The purpose of this invention is to provide a sepsis consultation method and system based on multi-turn dialogue, so as to solve at least one of the technical problems existing in the above-mentioned background art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a sepsis consultation method based on multi-turn dialogue, comprising:

[0008] Construct a two-dimensional sepsis knowledge graph, including a hierarchical clinical knowledge graph and an authoritative medical knowledge graph, to achieve multi-granular semantic modeling from macro-clinical scenarios to micro-medical concepts;

[0009] Based on the exploration-utilization reasoning mechanism, a multi-turn consultation dialogue architecture is designed. Coarse-grained task constraints are achieved through a structured sepsis diagnosis and treatment chain, and fine-grained reasoning generation control is achieved through a confidence-driven dynamic strategy during the diagnosis and treatment phase.

[0010] In the dialogue generation process, a cross-graph interactive hierarchical retrieval generation method is adopted. The authoritative medical knowledge graph is used as the semantic anchor point. The multi-level semantic subgraphs of the hierarchical clinical knowledge graph are activated step by step. Within the activated subgraphs, semantic field coupling matching nodes are used, and knowledge enhancement is carried out through a retrieval knowledge feedback mechanism.

[0011] Sepsis diagnosis and treatment suggestions are generated based on a large language model enhanced by multi-round consultation dialogue.

[0012] As a further limitation of the first aspect of the present invention, the two-dimensional sepsis knowledge graph includes a hierarchical clinical knowledge graph H-Clinical KG constructed based on sepsis clinical practice corpus and an authoritative standardized medical knowledge base Prime KG. The construction process of the hierarchical clinical knowledge graph is as follows:

[0013] For the input clinical text of sepsis Perform topic-driven segmentation to break down the text into... Semantic fragments focusing on a single theme Each segment focuses on a specific medical dimension;

[0014] LLM is used to extract entity relations and corresponding semantic descriptions from segmented semantic fragments to construct a basic knowledge graph. ,in For a collection of entities, Let be the set of relation edges. A semantic description set;

[0015] A multi-level clustering algorithm was used to hierarchically organize the basic knowledge graph to obtain a hierarchical clinical knowledge graph.

[0016] Among them, the authoritative medical knowledge graph Prime KG is based on multi-source standardized medical knowledge sources.

[0017] As a further limitation of the first aspect of the present invention, the exploration-exploitation reasoning mechanism comprises two collaborative control layers: an exploration layer and an exploitation layer.

[0018] The aforementioned exploration layer achieves coarse-grained constraints through a standardized sepsis diagnosis and treatment thought chain, which is strictly divided into three progressive stages:

[0019] Symptom assessment phase Given the original medical history input Collect the patient's basic information, infection symptoms, vital signs, and laboratory test results, combined with medical knowledge. Analyze the details of sepsis-related symptoms;

[0020] Infection diagnosis stage Prior knowledge based on the symptom assessment phase To classify the severity of sepsis and make inferences about the source of infection, determine whether the diagnosis criteria for sepsis are met, and identify potential sites of infection and types of pathogens;

[0021] Treatment decision stage Based on the diagnosis results and the patient's individual characteristics, a personalized sepsis treatment plan is developed.

[0022] As a further limitation of the first aspect of the present invention, the utilization layer implements fine-grained inference control based on confidence assessment, including:

[0023] Medical problems are categorized and category labels are determined; considering three dimensions—knowledge sufficiency, knowledge redundancy, and knowledge consistency—the confidence level of the current reasoning state is evaluated in real time at each reasoning stage.

[0024] Based on confidence assessment results Three dynamic action strategies are adopted:

[0025] High confidence strategy: when Exceeding the threshold and the problem category label Compared with the current stage Match or other categories At this time, the system selects a global termination action. It directly generates final treatment recommendations;

[0026] Medium confidence strategy: when Exceeding the threshold but the problem category label Compared with the current stage Mismatch and not of other categories When the time comes, select the stage to advance, update the status, enter the next stage, and perform detailed queries.

[0027] Low confidence strategy: when If the value is below the threshold, select a specific query refinement action within the selected stage and use a large language model to synthesize the original problem. Prior knowledge of dialogue and the current stage Generate targeted subqueries and produce new, enhanced search answers. Update status.

[0028] As a further limitation of the first aspect of the present invention, the cross-graph interactive hierarchical retrieval method includes three stages: semantic anchor point generation, anchor point-guided hierarchical retrieval, and retrieval knowledge feedback.

[0029] The semantic anchor generation process is as follows: Given a sepsis consultation query... First, semantic retrieval is performed on the triple set of the authoritative medical knowledge graph Prime KG to obtain a highly relevant subset of triples;

[0030] The anchor-guided hierarchical retrieval process is divided into two stages:

[0031] Candidate subgraph selection stage: The anchor set is mapped to the top-level aggregation node of the hierarchical clinical knowledge graph H-Clinical KG, and its semantic similarity with the anchor set is calculated; subgraphs with similarity greater than the threshold are activated, and then the middle and bottom subgraphs are activated step by step to form a multi-granularity candidate subgraph set.

[0032] Semantic field coupling matching stage: In the candidate subgraph, the semantic field coupling degree (SFCS) is used to quantify the comprehensive semantic association between nodes and anchors;

[0033] The aforementioned knowledge feedback mechanism utilizes the first The retrieval results of each round are used to construct an enhanced triplet set and extract an enhanced anchor set. This process is repeated iteratively until no new anchors are generated or the preset maximum number of iterations is reached.

[0034] As a further limitation of the first aspect of the present invention, the original medical consultation query Multiple rounds of medical consultation dialogue And by combining the prompt template with the input large language model, suggestions for the diagnosis and treatment of sepsis are generated:

[0035] .

[0036] Secondly, the present invention provides a sepsis consultation system based on multi-turn dialogue, comprising:

[0037] The module is used to construct a two-dimensional sepsis knowledge graph, including a hierarchical clinical knowledge graph and an authoritative medical knowledge graph, to achieve multi-granular semantic modeling from macro-clinical scenarios to micro-medical concepts;

[0038] The constraint module is used to design a multi-turn consultation dialogue architecture based on the exploration-utilization reasoning mechanism. It achieves coarse-grained task constraints through a structured sepsis diagnosis and treatment chain, and achieves fine-grained reasoning generation control through a confidence-driven dynamic strategy during the diagnosis and treatment phase.

[0039] The coupling matching module is used to adopt a cross-graph interactive hierarchical retrieval generation method in the dialogue generation process. It uses an authoritative medical knowledge graph as a semantic anchor point, activates multi-level semantic subgraphs of the hierarchical clinical knowledge graph step by step, uses semantic field coupling matching nodes in the activated subgraphs, and enhances knowledge through a retrieval knowledge feedback mechanism.

[0040] The generation module is used to generate sepsis diagnosis and treatment suggestions based on a large language model enhanced by multi-turn consultation dialogues.

[0041] Thirdly, the present invention provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the sepsis consultation method based on multi-turn dialogue as described in the first aspect.

[0042] Fourthly, the present invention provides a computer device including a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor invokes the program instructions to execute the sepsis consultation method based on multi-turn dialogue as described in the first aspect.

[0043] Fifthly, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions to implement the sepsis consultation method based on multi-turn dialogue as described in the first aspect.

[0044] The beneficial effects of this invention are as follows: By constructing a two-dimensional knowledge graph and performing cross-graph interactive retrieval, it effectively integrates standardized medical knowledge with personalized clinical experience, significantly improving the accuracy and reliability of sepsis diagnosis and treatment recommendations; the proposed exploration-utilization reasoning mechanism, through structured sepsis diagnosis and treatment chain constraints and a confidence-driven dynamic strategy, systematically alleviates the illusion problem of large language models in complex clinical reasoning, ensuring that the generated content conforms to the norms of clinical diagnosis and treatment logic; the proposed cross-graph interactive hierarchical retrieval method, through semantic anchor guidance, hierarchical activation, and knowledge feedback mechanisms, achieves multi-granular semantic association from macro-clinical scenarios to micro-medical concepts, balancing retrieval accuracy and knowledge coverage; the confidence evaluation system comprehensively considers knowledge sufficiency, redundancy, and consistency, adaptively balancing knowledge utilization and information exploration, improving the reasoning efficiency and diagnosis and treatment quality of multi-turn dialogues.

[0045] The advantages of additional aspects of the invention will be set forth more clearly in the following description or will be learned by practice of the invention. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart illustrating the framework of the sepsis consultation method based on multi-turn dialogue as described in an embodiment of the present invention. Detailed Implementation

[0048] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0049] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0050] It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as here.

[0051] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or groups thereof.

[0052] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0053] To facilitate understanding of the present invention, the present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0054] Those skilled in the art should understand that the accompanying drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily essential for implementing the present invention.

[0055] Example 1

[0056] In this embodiment 1, a sepsis consultation system based on multi-turn dialogue is first provided, including: a construction module for constructing a two-dimensional sepsis knowledge graph, including a hierarchical clinical knowledge graph and an authoritative medical knowledge graph, to achieve multi-granular semantic modeling from macro-clinical scenarios to micro-medical concepts; a constraint module for designing a multi-turn consultation dialogue architecture based on an exploration-utilization reasoning mechanism, implementing coarse-grained task constraints through a structured sepsis diagnosis and treatment chain, and achieving fine-grained reasoning generation control through a confidence-driven dynamic strategy during the diagnosis and treatment phase; a coupling and matching module for using a cross-graph interactive hierarchical retrieval generation method during the dialogue generation process, using the authoritative medical knowledge graph as a semantic anchor, activating multi-level semantic subgraphs of the hierarchical clinical knowledge graph level by level, using semantic field coupling and matching nodes within the activated subgraphs, and enhancing knowledge through a retrieval knowledge feedback mechanism; and a generation module for generating sepsis diagnosis and treatment suggestions based on a large language model enhanced by multi-turn consultation dialogue.

[0057] In this embodiment, the above-described system is used to implement a sepsis consultation method based on multi-turn dialogue, including the following steps:

[0058] (1) Construct a two-dimensional sepsis knowledge graph, including a hierarchical clinical knowledge graph and an authoritative medical knowledge graph, to achieve multi-granular semantic modeling from macro-clinical scenarios to micro-medical concepts;

[0059] (2) Based on the exploration-utilization reasoning mechanism, a multi-round consultation dialogue architecture is designed. Coarse-grained task constraints are achieved through a structured sepsis diagnosis and treatment chain, and fine-grained reasoning generation control is achieved through a confidence-driven dynamic strategy during the diagnosis and treatment phase.

[0060] (3) In the dialogue generation process, a cross-graph interactive hierarchical retrieval generation method is adopted. The authoritative medical knowledge graph is used as the semantic anchor point. The multi-level semantic subgraphs of the hierarchical clinical knowledge graph are activated step by step. Semantic field coupling matching nodes are used in the activated subgraphs, and knowledge enhancement is carried out through the retrieval knowledge feedback mechanism.

[0061] (4) Based on multi-round consultation dialogue, enhance the large language model to generate sepsis diagnosis and treatment suggestions, and construct the sepsis consultation device.

[0062] In step (1), the two-dimensional sepsis knowledge graph includes a hierarchical clinical knowledge graph H-Clinical KG constructed based on sepsis clinical practice corpus and an authoritative standardized medical knowledge base Prime KG. The construction process of the hierarchical clinical knowledge graph is as follows:

[0063] First, the input sepsis clinical text Perform topic-driven segmentation to break down the text into... Semantic fragments focusing on a single theme Each segment focuses on a specific medical dimension, including basic patient information, infection symptoms, organ function indicators, inflammatory markers, hemodynamic parameters, etc.

[0064] Secondly, LLM is used to extract entity relations and corresponding semantic descriptions from the segmented semantic fragments to construct a basic knowledge graph. ,in For a collection of entities, Let be the set of relation edges. A semantic description set;

[0065] Furthermore, a multi-level clustering algorithm is used to hierarchically organize the basic knowledge graph, resulting in a hierarchical clinical knowledge graph:

[0066]

[0067] in, For the number of levels, For the first The set of subgraphs of a layer This indicates the first layer. Sub-image, These are, respectively, the entity set, relation edge set, semantic description set, and subgraph semantic title.

[0068] The authoritative medical knowledge graph Prime KG is based on multi-source standardized medical knowledge sources and is formally defined as follows:

[0069] ;

[0070] in, It is a standardized set of medical entities, including sepsis-related diseases, symptoms, drugs, and test indicators. This is a set of triples that describe the medical relationships between entities; It is an authoritative definition at the entity level, including diagnostic criteria and treatment guidelines for sepsis.

[0071] In step (2), the exploration-exploitation reasoning mechanism comprises two collaborative control layers: the exploration layer and the exploitation layer.

[0072] The aforementioned exploration layer achieves coarse-grained constraints through a standardized sepsis diagnosis and treatment thought chain, which is strictly divided into three progressive stages:

[0073] Symptom assessment phase Given the original medical history input Collect the patient's basic information, infection symptoms, vital signs, and laboratory test results, combined with medical knowledge. Analyze the details of sepsis-related symptoms, including the location of the infection focus and SOFA score.

[0074] Infection diagnosis stage Prior knowledge based on the symptom assessment phase To classify the severity of sepsis and make inferences about the source of infection, determine whether the diagnosis criteria for sepsis are met, and identify potential sites of infection and types of pathogens.

[0075] Treatment decision stage Based on the diagnosis and individual patient characteristics, a personalized sepsis treatment plan is developed, including early fluid resuscitation, antibiotic selection, use of vasoactive drugs, and organ function support.

[0076] The aforementioned utilization layer implements fine-grained inference control based on confidence assessment. System maintenance status representation. :

[0077] ,

[0078] in, This is the original medical history input. Indicates the current stage of the thought process;

[0079] for Individual queries - collections of medical dialogue knowledge with correct answers.

[0080] The system first categorizes the input questions, using category labels. The definition is as follows:

[0081] ;

[0082] in, These correspond to the three stages of symptom assessment, infection diagnosis, and treatment decision-making, respectively. Indicates other categories.

[0083] The system evaluates the confidence level of the current inference state in real time at each inference stage. Taking into account three dimensions: knowledge sufficiency, knowledge redundancy, and knowledge consistency:

[0084] ,

[0085] in, , These are weighted parameters. Knowledge sufficiency measures the current knowledge in medical dialogue. For the original problem Coverage completeness; knowledge redundancy measurement of generated content With Dialogue Knowledge Repetition rate; knowledge consistency measures generated content and Prime KG sepsis definition Consistency.

[0086] Based on confidence assessment results The system employs three dynamic action strategies:

[0087] High confidence strategy: when Exceeding the threshold and the problem category label Compared with the current stage Match or other categories At this time, the system selects a global termination action. It directly generates final treatment recommendations;

[0088] Medium confidence strategy: when Exceeding the threshold but the problem category label Compared with the current stage Mismatch and not of other categories At that time, the system selects a phase to advance the action. The status has been updated to ,

[0089] Proceed to the next stage and refine the query;

[0090] Low confidence strategy: when If the value is below the threshold, the system will select a phase for refined query actions. Using a large language model to synthesize the original problem Prior knowledge of dialogue and the current stage Generate targeted subqueries :

[0091] ;

[0092] And generate new enhanced search answers The status has been updated to:

[0093] .

[0094] In step (3), the cross-graph interactive hierarchical retrieval method includes three stages: semantic anchor generation, anchor-guided hierarchical retrieval, and retrieval knowledge feedback.

[0095] The semantic anchor generation process is as follows: Given a sepsis consultation query... First, in the triplet set of the authoritative medical knowledge graph Prime KG Semantic retrieval is performed to obtain a highly relevant subset of triples:

[0096] ,

[0097] in, This is the similarity threshold. Furthermore, from... Extract the entity set as the semantic anchor set:

[0098] ,

[0099] The anchor-guided hierarchical retrieval process is divided into two stages:

[0100] Candidate subgraph selection stage: The anchor point set Mapped to the top-level aggregation node of the hierarchical clinical knowledge graph H-Clinical KG Calculate its semantic similarity with the anchor set:

[0101] ,

[0102] Activate subgraphs with similarity greater than a threshold, and then activate the middle and bottom subgraphs step by step to form a multi-granularity candidate subgraph set;

[0103] Semantic Field Coupling Matching Stage: In the candidate subgraph, the semantic field coupling degree (SFCS) is used to quantify the comprehensive semantic association between nodes and anchors.

[0104] ;

[0105] in, This is the balance coefficient; It is a node The candidate subgraph's intra-neighbor set. Based on SFCS values, the Top-K nodes and their associated medical texts are selected. As the first Search results for round .

[0106] The aforementioned knowledge feedback mechanism utilizes the first Search results for round Construct enhanced triple set :

[0107] ,

[0108] And extract the enhanced anchor point set :

[0109] ,

[0110] This process is repeated iteratively until no new anchor points are generated or the preset maximum number of iterations is reached. Finally, the query will be performed. Knowledge obtained through multiple rounds of retrieval With Prime KG Triple The answer is obtained by combining input LLM and reasoning:

[0111] In step (4), the original medical history query is... Multiple rounds of medical consultation dialogue And by combining the prompt template with the input large language model, suggestions for the diagnosis and treatment of sepsis are generated: .

[0112] Example 2

[0113] In this embodiment 2, a sepsis consultation device based on multi-turn dialogue (MEERAG) is provided to solve the problems of medical reasoning illusion, insufficient accuracy of knowledge retrieval, and weak clinical context modeling ability in the prior art.

[0114] like Figure 1 As shown, the sepsis consultation device described in this embodiment implements a sepsis consultation method based on multi-turn dialogue, specifically including the following process steps:

[0115] (1) Construct a two-dimensional sepsis knowledge graph, including a hierarchical clinical knowledge graph and a standard medical knowledge graph, to achieve multi-granular semantic modeling from macro-clinical scenarios to micro-medical concepts;

[0116] (2) Based on the exploration-utilization reasoning mechanism, a multi-round consultation dialogue architecture is designed. Coarse-grained task constraints are achieved through a structured sepsis diagnosis and treatment chain, and fine-grained reasoning generation control is achieved through a confidence-driven dynamic strategy during the diagnosis and treatment phase.

[0117] (3) In the dialogue generation process, a cross-graph interactive hierarchical retrieval generation method is adopted. The authoritative medical knowledge graph is used as the semantic anchor point. The multi-level semantic subgraphs of the hierarchical clinical knowledge graph are activated step by step. Semantic field coupling matching nodes are used in the activated subgraphs, and knowledge enhancement is carried out through the retrieval knowledge feedback mechanism.

[0118] (4) Based on multi-round consultation dialogue, enhance the large language model to generate sepsis diagnosis and treatment suggestions, and construct the sepsis consultation device.

[0119] In this embodiment, for step (1), a two-dimensional sepsis knowledge graph is constructed, including a hierarchical clinical knowledge graph and a standard medical knowledge graph.

[0120] In step (1), the two-dimensional sepsis knowledge graph includes a hierarchical clinical knowledge graph H-Clinical KG constructed based on sepsis clinical practice corpus and an authoritative standardized medical knowledge base Prime KG. The construction process of the hierarchical clinical knowledge graph is as follows:

[0121] First, the input sepsis clinical text Perform topic-driven segmentation to break down the text into... Semantic fragments focusing on a single theme Each segment focuses on a specific medical dimension, including basic patient information, infection symptoms, organ function indicators, inflammatory markers, hemodynamic parameters, etc.

[0122] Secondly, LLM is used to extract entity relations and corresponding semantic descriptions from the segmented semantic fragments to construct a basic knowledge graph. ,in For a collection of entities, Let be the set of relation edges. A semantic description set;

[0123] Furthermore, a multi-level clustering algorithm is used to hierarchically organize the basic knowledge graph, resulting in a hierarchical clinical knowledge graph:

[0124] ;

[0125] in, For the number of levels, For the first The set of subgraphs of a layer This indicates the first layer. Sub-image, These are, respectively, the entity set, relation edge set, semantic description set, and subgraph semantic title.

[0126] The aforementioned standard medical knowledge graph, Prime KG, is based on multi-source standardized medical knowledge sources and is formally defined as follows:

[0127] ;

[0128] in, It is a standardized set of medical entities, including sepsis-related diseases, symptoms, drugs, and test indicators. This is a set of triples that describe the medical relationships between entities; It is an authoritative definition at the entity level, including diagnostic criteria and treatment guidelines for sepsis.

[0129] In this embodiment, the multi-turn consultation dialogue architecture specifically includes the following:

[0130] (I) Constraints of the Thinking Chain in the Diagnosis and Treatment of Sepsis:

[0131] In step (2), the exploration-exploitation reasoning mechanism comprises two collaborative control layers: the exploration layer and the exploitation layer.

[0132] The aforementioned exploration layer achieves coarse-grained constraints through a standardized sepsis diagnosis and treatment thought chain, which is strictly divided into three progressive stages:

[0133] Symptom assessment phase Given the original medical history input Collect the patient's basic information, infection symptoms, vital signs, and laboratory test results, combined with medical knowledge. Analyze the details of sepsis-related symptoms, including the location of the infection focus and SOFA score.

[0134] Infection diagnosis stage Prior knowledge based on the symptom assessment phase To classify the severity of sepsis and make inferences about the source of infection, determine whether the diagnosis criteria for sepsis are met, and identify potential sites of infection and types of pathogens.

[0135] Treatment decision stage Based on the diagnosis and individual patient characteristics, a personalized sepsis treatment plan is developed, including early fluid resuscitation, antibiotic selection, use of vasoactive drugs, and organ function support.

[0136] (ii) Confidence-based inference control layer:

[0137] The aforementioned utilization layer implements fine-grained inference control based on confidence assessment. System maintenance status representation. : ,

[0138] in, This is the original medical history input. Indicates the current stage of the thought process;

[0139] for Individual queries - collections of medical dialogue knowledge with correct answers.

[0140] The system first categorizes the input questions, using category labels. The definition is as follows:

[0141] ;

[0142] in, These correspond to the three stages of symptom assessment, infection diagnosis, and treatment decision-making, respectively. Indicates other categories.

[0143] The system evaluates the confidence level of the current inference state in real time at each inference stage. Taking into account three dimensions: knowledge sufficiency, knowledge redundancy, and knowledge consistency:

[0144] ,

[0145] in, , These are weighted parameters. Knowledge sufficiency measures the current knowledge in medical dialogue. For the original problem Coverage completeness; knowledge redundancy measurement of generated content With Dialogue Knowledge Repetition rate; knowledge consistency measures generated content and Prime KG sepsis definition Consistency.

[0146] Based on confidence assessment results The system employs three dynamic action strategies:

[0147] High confidence strategy: when Exceeding the threshold and the problem category label Compared with the current stage Match or other categories At this time, the system selects a global termination action. It directly generates final treatment recommendations;

[0148] Medium confidence strategy: when Exceeding the threshold but the problem category label Compared with the current stage Mismatch and not of other categories At that time, the system selects a phase to advance the action. The status has been updated to Proceed to the next stage and refine the query;

[0149] Low confidence strategy: when If the value is below the threshold, the system will select a phase for refined query actions. Using a large language model to synthesize the original problem Prior knowledge of dialogue and the current stage Generate targeted subqueries : ;

[0150] And generate new enhanced search answers The status has been updated to .

[0151] (III) Enhanced Generation Through Cross-Graph Interactive Retrieval:

[0152] In step (3), the cross-graph interactive hierarchical retrieval method includes three stages: semantic anchor generation, anchor-guided hierarchical retrieval, and retrieval knowledge feedback.

[0153] The semantic anchor point generation process is as follows:

[0154] Given sepsis consultation query First, in the triplet set of the authoritative medical knowledge graph Prime KG Semantic retrieval is performed to obtain a highly relevant subset of triples:

[0155] ;

[0156] in, This is the similarity threshold. Furthermore, from... Extract the entity set as the semantic anchor set:

[0157] .

[0158] The anchor-guided hierarchical retrieval process is divided into two stages:

[0159] Candidate subgraph selection stage: The anchor point set Mapped to the top-level aggregation node of the hierarchical clinical knowledge graph H-Clinical KG Calculate its semantic similarity with the anchor set:

[0160] ,

[0161] Activate subgraphs with similarity greater than a threshold, and then activate the middle and bottom subgraphs step by step to form a multi-granularity candidate subgraph set;

[0162] Semantic Field Coupling Matching Stage: In the candidate subgraph, the semantic field coupling degree (SFCS) is used to quantify the comprehensive semantic association between nodes and anchors.

[0163] ;

[0164] in, This is the balance coefficient; It is a node The candidate subgraph's intra-neighbor set. Based on SFCS values, the Top-K nodes and their associated medical texts are selected. As the first Search results for round .

[0165] The aforementioned knowledge feedback mechanism utilizes the first Search results for round Construct enhanced triple set :

[0166] ;

[0167] And extract the enhanced anchor point set :

[0168] ,

[0169] This process is repeated iteratively until no new anchor points are generated or the preset maximum number of iterations is reached. Finally, the query will be performed. Knowledge obtained through multiple rounds of retrieval With Prime KG Triple The answer is obtained by combining input LLM and reasoning: .

[0170] (iv) Generation of sepsis diagnosis and treatment recommendations

[0171] In step (4), the original medical history query is... Multiple rounds of medical consultation dialogue And by combining the prompt template with the input large language model, suggestions for the diagnosis and treatment of sepsis are generated: .

[0172] In this embodiment, the sepsis consultation method described in this embodiment is compared with the traditional large language model generation method in terms of consultation result examples. The results are as follows:

[0173] The entered medical question is as follows: A 34-year-old female patient was admitted to the intensive care unit due to septic shock and bacteremia caused by a urinary tract infection. Three days after admission, despite adequate fluid resuscitation and high-dose vasopressors, she continues to experience hypotension. Blood cultures are positive for Escherichia coli. She has been receiving targeted antibiotic treatment since admission. The patient has no history of major illness and denies using any prohibited medications. Current medications include norepinephrine, ceftriaxone, and acetaminophen. The patient's general condition is fair; her temperature is 37.5°C, heart rate is 96 bpm, and blood pressure is 85 / 55 mmHg. Physical examination of the back reveals tenderness at both costovertebral angles. Thyroid examination is unremarkable. Laboratory results showed: On the first day of admission, white blood cell count was 18500 / mm3, hemoglobin was 14.1 g / dl, serum creatinine was 1.4 mg / dl, and fasting blood glucose was 95 mg / dl; on the third day of admission, white blood cell count was 10300 / mm3, hemoglobin was 13.4 g / dl, serum creatinine was 0.9 mg / dl, fasting blood glucose was 100 mg / dl, thyroid-stimulating hormone was 1.8 μU / ml, free triiodothyronine was 0.1 ng / dl, and free thyroid hormone was 0.9 ng / dl. Repeat blood cultures were negative. Chest X-ray was normal. What treatment should be given?

[0174] The recommended treatment plan output by the large language model is as follows: On the third day of ICU admission, the patient still presented with persistent hypotension. Although blood cultures were negative and infection markers (white blood cell count) decreased significantly, blood pressure remained at 85 / 55 mmHg, requiring norepinephrine support. Laboratory results showed low T3 and T4 levels and normal TSH, suggesting possible euthyroid sickness syndrome. This condition is common in critically ill patients and usually requires no specific treatment.

[0175] The recommended treatment plan output by the sepsis history taking method described in this embodiment is as follows: For this patient with persistent hypotension following septic shock, the most appropriate treatment is intravenous hydrocortisone 200 mg / day. The core issue for this patient is that although the primary infection has been effectively controlled (negative blood cultures, decreased white blood cell count, and improved creatinine), hypotension persists on day 3 of the ICU, requiring vasopressors. This clinical presentation strongly suggests critical illness-related corticosteroid insufficiency, a complication that may occur in 10-20% of septic shock patients, characterized by insufficient corticosteroid activity relative to disease severity. The patient's thyroid function tests showed low T3 and T4 levels but normal TSH, consistent with common non-thyroid syndromes in critically ill patients, requiring no special intervention. This further supports the diagnosis of critical illness-related corticosteroid insufficiency. According to the Save the Sepsis Campaign guidelines, hydrocortisone should be administered to septic shock patients who require vasopressors to maintain blood pressure despite adequate fluid resuscitation. This treatment can enhance the sensitivity of blood vessels to catecholamines, typically producing significant hemodynamic improvements within 24-48 hours, thereby reducing the dosage of vasopressors.

[0176] Example 3

[0177] This embodiment 3 provides a non-transitory computer-readable storage medium for storing computer instructions. When executed by a processor, the computer instructions implement the sepsis consultation method based on multi-turn dialogue as described above. The method includes:

[0178] Construct a two-dimensional sepsis knowledge graph, including a hierarchical clinical knowledge graph and an authoritative medical knowledge graph, to achieve multi-granular semantic modeling from macro-clinical scenarios to micro-medical concepts;

[0179] Based on the exploration-utilization reasoning mechanism, a multi-turn consultation dialogue architecture is designed. Coarse-grained task constraints are achieved through a structured sepsis diagnosis and treatment chain, and fine-grained reasoning generation control is achieved through a confidence-driven dynamic strategy during the diagnosis and treatment phase.

[0180] In the dialogue generation process, a cross-graph interactive hierarchical retrieval generation method is adopted. The authoritative medical knowledge graph is used as the semantic anchor point. The multi-level semantic subgraphs of the hierarchical clinical knowledge graph are activated step by step. Within the activated subgraphs, semantic field coupling matching nodes are used, and knowledge enhancement is carried out through a retrieval knowledge feedback mechanism.

[0181] Sepsis diagnosis and treatment suggestions are generated based on a large language model enhanced by multi-round consultation dialogue.

[0182] Example 4

[0183] This embodiment 4 provides a computer device, including a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the sepsis consultation method based on multi-turn dialogue as described above, the method including:

[0184] Construct a two-dimensional sepsis knowledge graph, including a hierarchical clinical knowledge graph and an authoritative medical knowledge graph, to achieve multi-granular semantic modeling from macro-clinical scenarios to micro-medical concepts;

[0185] Based on the exploration-utilization reasoning mechanism, a multi-turn consultation dialogue architecture is designed. Coarse-grained task constraints are achieved through a structured sepsis diagnosis and treatment chain, and fine-grained reasoning generation control is achieved through a confidence-driven dynamic strategy during the diagnosis and treatment phase.

[0186] In the dialogue generation process, a cross-graph interactive hierarchical retrieval generation method is adopted. The authoritative medical knowledge graph is used as the semantic anchor point. The multi-level semantic subgraphs of the hierarchical clinical knowledge graph are activated step by step. Within the activated subgraphs, semantic field coupling matching nodes are used, and knowledge enhancement is carried out through a retrieval knowledge feedback mechanism.

[0187] Sepsis diagnosis and treatment suggestions are generated based on a large language model enhanced by multi-round consultation dialogue.

[0188] Example 5

[0189] This embodiment 5 provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions to implement the sepsis consultation method based on multi-turn dialogue as described above, the method including:

[0190] Construct a two-dimensional sepsis knowledge graph, including a hierarchical clinical knowledge graph and an authoritative medical knowledge graph, to achieve multi-granular semantic modeling from macro-clinical scenarios to micro-medical concepts;

[0191] Based on the exploration-utilization reasoning mechanism, a multi-turn consultation dialogue architecture is designed. Coarse-grained task constraints are achieved through a structured sepsis diagnosis and treatment chain, and fine-grained reasoning generation control is achieved through a confidence-driven dynamic strategy during the diagnosis and treatment phase.

[0192] In the dialogue generation process, a cross-graph interactive hierarchical retrieval generation method is adopted. The authoritative medical knowledge graph is used as the semantic anchor point. The multi-level semantic subgraphs of the hierarchical clinical knowledge graph are activated step by step. Within the activated subgraphs, semantic field coupling matching nodes are used, and knowledge enhancement is carried out through a retrieval knowledge feedback mechanism.

[0193] Sepsis diagnosis and treatment suggestions are generated based on a large language model enhanced by multi-round consultation dialogue.

[0194] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0195] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0196] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0197] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0198] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solutions disclosed in the present invention, various modifications or variations that can be made by those skilled in the art without creative effort should be included within the scope of protection of the present invention.

Claims

1. A sepsis consultation method based on multi-turn dialogue, characterized in that, include: Construct a two-dimensional sepsis knowledge graph, including a hierarchical clinical knowledge graph and an authoritative medical knowledge graph, to achieve multi-granular semantic modeling from macro-clinical scenarios to micro-medical concepts; Based on the exploration-utilization reasoning mechanism, a multi-turn consultation dialogue architecture is designed. Coarse-grained task constraints are achieved through a structured sepsis diagnosis and treatment chain, and fine-grained reasoning generation control is achieved through a confidence-driven dynamic strategy during the diagnosis and treatment phase. In the dialogue generation process, a cross-graph interactive hierarchical retrieval generation method is adopted. The authoritative medical knowledge graph is used as the semantic anchor point. The multi-level semantic subgraphs of the hierarchical clinical knowledge graph are activated step by step. Within the activated subgraphs, semantic field coupling matching nodes are used, and knowledge enhancement is carried out through a retrieval knowledge feedback mechanism. Sepsis diagnosis and treatment suggestions are generated based on a large language model enhanced by multi-round consultation dialogue.

2. The sepsis consultation method based on multi-turn dialogue according to claim 1, characterized in that, The two-dimensional sepsis knowledge graph comprises a hierarchical clinical knowledge graph (H-Clinical KG) constructed based on sepsis clinical practice corpus and an authoritative standardized medical knowledge base (Prime KG). The construction process of the hierarchical clinical knowledge graph is as follows: For the input clinical text of sepsis Perform topic-driven segmentation to break down the text into... Semantic fragments focusing on a single theme Each segment focuses on a specific medical dimension; LLM is used to extract entity relations and corresponding semantic descriptions from segmented semantic fragments to construct a basic knowledge graph. ,in For a collection of entities, Let be the set of relation edges. A semantic description set; A multi-level clustering algorithm was used to hierarchically organize the basic knowledge graph to obtain a hierarchical clinical knowledge graph. Among them, the authoritative medical knowledge graph Prime KG is based on multi-source standardized medical knowledge sources.

3. The sepsis consultation method based on multi-turn dialogue according to claim 1, characterized in that, The exploration-exploitation reasoning mechanism comprises two collaborative control layers: the exploration layer and the exploitation layer. The aforementioned exploration layer achieves coarse-grained constraints through a standardized sepsis diagnosis and treatment thought chain, which is strictly divided into three progressive stages: Symptom assessment phase Given the original medical history input Collect the patient's basic information, infection symptoms, vital signs, and laboratory test results, combined with medical knowledge. Analyze the details of sepsis-related symptoms; Infection diagnosis stage Prior knowledge based on the symptom assessment phase To classify the severity of sepsis and make inferences about the source of infection, determine whether the diagnosis criteria for sepsis are met, and identify potential sites of infection and types of pathogens; Treatment decision stage Based on the diagnosis results and the patient's individual characteristics, a personalized sepsis treatment plan is developed.

4. The sepsis consultation method based on multi-turn dialogue according to claim 1, characterized in that, The aforementioned utilization layer implements fine-grained inference control based on confidence assessment, including: Medical problems are categorized and category labels are determined; considering three dimensions—knowledge sufficiency, knowledge redundancy, and knowledge consistency—the confidence level of the current reasoning state is evaluated in real time at each reasoning stage. Based on confidence assessment results Three dynamic action strategies are adopted: High confidence strategy: when Exceeding the threshold and the problem category label Compared with the current stage Match or other categories At this time, the system selects a global termination action. It directly generates final treatment recommendations; Medium confidence strategy: when Exceeding the threshold but the problem category label Compared with the current stage Mismatch and not of other categories When the time comes, select the stage to advance, update the status, enter the next stage, and perform detailed queries. Low confidence strategy: when If the value is below the threshold, select a specific query refinement action within the selected stage and use a large language model to synthesize the original problem. Prior knowledge of dialogue and the current stage Generate targeted subqueries and produce new, enhanced search answers. Update status.

5. The sepsis consultation method based on multi-turn dialogue according to claim 1, characterized in that, The cross-graph interactive hierarchical retrieval method comprises three stages: semantic anchor generation, anchor-guided hierarchical retrieval, and retrieval knowledge feedback. The semantic anchor generation process is as follows: Given a sepsis consultation query... First, semantic retrieval is performed on the triple set of the authoritative medical knowledge graph Prime KG to obtain a highly relevant subset of triples; The anchor-guided hierarchical retrieval process is divided into two stages: Candidate subgraph selection stage: The anchor set is mapped to the top-level aggregation node of the hierarchical clinical knowledge graph H-Clinical KG, and its semantic similarity with the anchor set is calculated; subgraphs with similarity greater than the threshold are activated, and then the middle and bottom subgraphs are activated step by step to form a multi-granularity candidate subgraph set. Semantic field coupling matching stage: In the candidate subgraph, the semantic field coupling degree (SFCS) is used to quantify the comprehensive semantic association between nodes and anchors; The aforementioned knowledge feedback mechanism utilizes the first The retrieval results of each round are used to construct an enhanced triplet set and extract an enhanced anchor set. This process is repeated iteratively until no new anchors are generated or the preset maximum number of iterations is reached.

6. The sepsis consultation method based on multi-turn dialogue according to claim 1, characterized in that, Original medical consultation query Multiple rounds of medical consultation dialogue And by combining the prompt template with the input large language model, suggestions for the diagnosis and treatment of sepsis are generated: 。 7. A sepsis consultation system based on multi-turn dialogue, characterized in that, include: The module is used to construct a two-dimensional sepsis knowledge graph, including a hierarchical clinical knowledge graph and an authoritative medical knowledge graph, to achieve multi-granular semantic modeling from macro-clinical scenarios to micro-medical concepts; The constraint module is used to design a multi-turn consultation dialogue architecture based on the exploration-utilization reasoning mechanism. It achieves coarse-grained task constraints through a structured sepsis diagnosis and treatment chain, and achieves fine-grained reasoning generation control through a confidence-driven dynamic strategy during the diagnosis and treatment phase. The coupling matching module is used to adopt a cross-graph interactive hierarchical retrieval generation method in the dialogue generation process. It uses an authoritative medical knowledge graph as a semantic anchor point, activates multi-level semantic subgraphs of the hierarchical clinical knowledge graph step by step, uses semantic field coupling matching nodes in the activated subgraphs, and enhances knowledge through a retrieval knowledge feedback mechanism. The generation module is used to generate sepsis diagnosis and treatment suggestions based on a large language model enhanced by multi-turn consultation dialogues.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the sepsis consultation method based on multi-turn dialogue as described in any one of claims 1-6.

9. A computer device, characterized in that, The method includes a memory and a processor, the processor and the memory communicating with each other, the memory storing program instructions that can be executed by the processor, and the processor calling the program instructions to execute the sepsis consultation method based on multi-turn dialogue as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions to implement the sepsis consultation method based on multi-turn dialogue as described in any one of claims 1-6.

Images