A human cognition-inspired hypergraph clinical decision framework construction method

Abstract

This invention discloses a method for constructing a hypergraph clinical decision-making framework inspired by human cognition, relating to the fields of medical artificial intelligence and clinical decision support technology. Specifically, this method first extracts conceptual entities from medical knowledge, dividing them into condition and decision sets, and discretizing continuous variables. Second, under logical constraints, it combines conditional concepts to construct production reasoning rules, represented as hyperedges. Subsequently, concepts are used as hypergraph nodes, combined with hyperedges to construct the hypergraph decision structure. Finally, the acquired patient profile information is matched with conditional nodes to determine activated hyperedges for rule filtering, thereby generating candidate decisions and outputting their corresponding reasoning paths. This invention supports parallel modeling of multi-source medical guidelines, realizes multi-condition joint reasoning and full transparency in complex clinical decisions, and significantly improves the reliability of decision support.

Description

Technical Field

[0001] This invention relates to the field of medical artificial intelligence and clinical decision support technology, specifically to a method for constructing a hypergraph clinical decision framework inspired by human cognition. Background Technology

[0002] The statements in this section are provided only as background information in relation to this disclosure and may not constitute prior art.

[0003] Clinical decision-making is a core aspect of medical practice, and its quality directly impacts the accuracy of disease diagnosis, the rationality of treatment plans, and the improvement of patient prognosis. In actual clinical work, physicians need to synthesize patient complaints, medical records, physical examination results, laboratory test indicators, and medical imaging data, while also incorporating clinical guidelines, expert consensus, and personal experience to make complex judgments and choices within a limited timeframe. This process is characterized by diverse information sources, complex decision-making conditions, and implicit reasoning paths, placing high demands on physicians' professional abilities and cognitive load.

[0004] With the continuous growth of medical knowledge and the increasing complexity of clinical scenarios, traditional decision-making methods relying on human experience are no longer sufficient to meet the demands for efficient, standardized, and consistent diagnosis and treatment. To assist clinical decision-making, existing technologies are gradually introducing rule-based expert systems and intelligent decision-making models based on machine learning and deep learning. Among them, expert systems typically achieve diagnostic reasoning by manually constructing rule bases. Although they have a certain degree of interpretability, rule construction is costly, has poor scalability, and is difficult to cope with complex and ever-changing clinical situations. While large-scale model methods based on deep learning have shown strong predictive capabilities in various medical tasks, their internal decision-making processes exhibit obvious "black box" characteristics. It is difficult to clearly explain how the model selects and weighs different medical factors in specific diagnostic and treatment decisions, thus limiting its reliable application in clinical scenarios.

[0005] Furthermore, existing medical knowledge representation methods mostly employ knowledge graph structures, which primarily describe the relationships between medical entities in the form of binary relations. This makes it difficult to effectively express the joint relationships involving multiple conditional concepts in decision-making. In actual clinical decision-making, the formulation of disease diagnosis and treatment plans often relies on a comprehensive judgment of multiple medical indicators, symptoms, signs, and background factors. Simple pairwise relation modeling cannot accurately reflect the true clinical reasoning logic. Simultaneously, different clinical guidelines or expert consensuses differ in diagnostic thresholds, grading standards, and treatment strategies. Existing technologies lack a unified, comparable, and adjustable structured representation method to support parallel modeling across multiple guidelines and personalized decision-making.

[0006] Therefore, there is an urgent need for a clinical decision-making framework that can simulate the slow thinking process in human cognition, explicitly characterize multi-condition joint reasoning relationships, and possess good interpretability and scalability, so as to achieve the structured expression of complex medical knowledge and transparent support for the clinical decision-making process, thereby improving the reliability and practical value of clinical decision support systems in practical applications. Summary of the Invention

[0007] The purpose of this invention is to propose a method for constructing a hypergraph-based clinical decision-making framework inspired by human cognition. By introducing hypergraph structures and production reasoning rules, it explicitly models multi-source medical concepts and their joint relationships, simulates the slow-thinking clinical decision-making process in human cognition, and realizes the structured expression and interpretable reasoning of complex clinical decision-making logic, thereby improving the transparency, credibility and practical application value of clinical decision support systems.

[0008] The technical solution of the present invention is as follows: A method for constructing a hypergraph-based clinical decision-making framework inspired by human cognition, comprising: Based on the clinical application scenario of the target disease, determine the type of clinical decision-making task and collect medical knowledge related to the clinical decision-making task; Extract medical concept entities related to the clinical decision-making task from the medical knowledge. The medical concept entities include symptoms and signs, examination indicators, disease status and treatment plan. Divide the medical concept entities into conditional concepts that characterize the patient status and decisional concepts that characterize the diagnosis or treatment, so as to form conditional set and decisional set respectively. Define the value space of the medical concept entity; when the value space of the medical concept entity is represented as a continuous medical variable, the continuous medical variable is discretized according to the corresponding preset threshold and transformed into a decidable logical condition; Based on preset logical constraints, multiple conditional concepts in the condition set are combined to construct production reasoning rules, and the production reasoning rules are represented as hyperedges. All conditional concepts in the condition set are used as conditional nodes, and all decisional concepts in the decision set are used as decisional nodes. The conditional nodes and the decisional nodes are used together as hypergraph nodes. The hyperedges are combined with the hypergraph nodes to construct a hypergraph decision structure. Obtain patient profile information at the current treatment time point, and match the patient profile information with the condition nodes in the hypergraph decision structure; Based on the matching results between the patient profile information and the condition nodes, the activated hyperedges are determined to filter out the activated production reasoning rules. The corresponding candidate decision set is generated according to the decision nodes pointed to by the activated production reasoning rules, and the candidate decision set and its corresponding reasoning path are output to achieve an interpretable presentation of the clinical decision-making process.

[0009] Furthermore, the clinical decision-making task types include at least one of diagnostic decision-making, treatment decision-making, decision-making to obtain further information, and decision-making to delay intervention.

[0010] Furthermore, the medical knowledge mentioned is derived from at least one of clinical guidelines, expert consensus, evidence-based medicine literature, and clinical practice experience.

[0011] Furthermore, the value space of the medical concept entity includes a numerical value space, a Boolean value space, and an enumerated value space.

[0012] Furthermore, when the value space of the medical concept entity is represented as a continuous medical variable, the continuous medical variable is discretized according to the corresponding preset threshold and transformed into a decidable logical condition, including: When the value space of the medical concept entity is the numerical value space and is represented as the continuous medical variable, the continuous medical variable is compared with the preset threshold and mapped to a discrete logical value that satisfies or does not satisfy the determinable logical condition.

[0013] Furthermore, the production reasoning rules adopt the "if-then" rule form; The antecedent of the rule is composed of at least one combination of the multiple conditional concepts based on logical conjunction, logical disjunction, or logical negation, and the consequent of the rule is composed of one or more of the decision concepts; the preset logical constraints include logical conjunction, logical disjunction, logical negation, and implication.

[0014] Furthermore, each of the production reasoning rules corresponds to a hyperedge, which simultaneously connects multiple condition nodes and points to at least one decision node.

[0015] Furthermore, the hypergraph decision structure also includes global system information for storing the meaning of medical concepts, the source of thresholds, and literature references.

[0016] Furthermore, obtaining the patient profile information at the current treatment time point includes: The patient profile information is constructed by acquiring at least one of the patient's chief complaint information, consultation record, physical examination results, laboratory test results, and medical imaging information.

[0017] Furthermore, the hypergraph decision structure supports parallel modeling of medical knowledge from different sources under the same structured representation system; The process of selecting activated production reasoning rules includes: dynamically activating corresponding rule combinations from production reasoning rules corresponding to medical knowledge from different sources, based on the patient profile information.

[0018] Compared with existing technologies, the advantages of this invention are: This invention models the joint relationships of multiple medical concepts using a hypergraph structure and combines production reasoning rules with first-order predicate logic to achieve explicit expression and traceable reasoning of complex clinical decision-making processes. Compared with existing decision-making methods based on experience or black-box models, this invention can intuitively present the role and logical relationship of different medical factors in decision-making, significantly improving the interpretability and credibility of clinical decision support systems. At the same time, this framework supports parallel modeling and flexible configuration of different clinical guidelines and expert rules under a unified structure, which helps reduce the cognitive load of clinical decision-making, improve diagnostic and treatment efficiency, and provide reliable structured support for individualized diagnosis and treatment in complex clinical scenarios. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments recorded in the embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0020] Figure 1 This is a schematic diagram of the construction steps of the Hypergraph clinical decision framework provided by the present invention: Figure 2 This is a schematic diagram of the reasoning rules consisting of the condition set and the decision set in the clinical decision-making process provided by the present invention: Figure 3 This invention provides an overview of a hypergraph clinical decision-making structure built upon a specific medical guideline. Figure 4 This invention provides a hyperboundary decision structure for peri-implant health diagnostic criteria. Figure 5 This is a schematic diagram comparing the same hyperedge rules for different medical knowledge using the hypergraph clinical decision framework provided by this invention. Figure 6 This is a visualization of the effect of the Hypergraph Clinical Decision-Making Framework provided by this invention on complex clinical decision-making processes. Detailed Implementation

[0021] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do 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 said element.

[0022] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0023] Example 1 First and foremost, it must be clearly stated that the human cognition-inspired hypergraph clinical decision-making framework construction method provided in this embodiment is essentially a data processing and logical operation method based on computers and underlying algorithms. Each step of this method relies on electronic computing devices with data processing capabilities (such as medical auxiliary diagnostic servers, cloud computing platforms, or local computer terminals) to automate the execution of underlying code. The candidate decisions and inference paths output by this method are only intended to provide clinicians with multi-dimensional structured auxiliary reference information to reduce their cognitive load, and are not intended to replace doctors in making direct disease diagnoses or treatment conclusions. Therefore, the solution described in this embodiment does not fall under the categories of methods for diagnosing and treating diseases as stipulated in Article 25 of the Patent Law, and is therefore subject to patent protection.

[0024] Specifically, such as Figure 1 As shown, this embodiment provides a method for constructing a hypergraph-based clinical decision-making framework inspired by human cognition. Its overall process and core computational logic are as follows: S101: Based on the clinical application scenario of the target disease, determine the type of clinical decision-making task and collect medical knowledge related to the clinical decision-making task.

[0025] Specifically, the server first needs to establish the clinical scenario that the system is currently dealing with. The types of clinical decision-making tasks include at least one of the following: diagnostic decision, treatment decision, decision to obtain further information, and decision to delay intervention.

[0026] To construct a high-confidence underlying decision-making logic, the medical knowledge is derived from at least one of clinical guidelines, expert consensus, evidence-based medicine literature, and clinical practice experience. In practice, medical knowledge can be initially collected and digitized from the aforementioned multi-source unstructured text through natural language processing (NLP) technology or manual annotation.

[0027] S102: Extract medical concept entities related to the clinical decision-making task from the medical knowledge. The medical concept entities include symptoms and signs, examination indicators, disease status and treatment plan. The medical concept entities are symbolically represented and divided into conditional concepts used to characterize the patient's status and decisional concepts used to characterize the diagnosis or treatment, so as to form a condition set and a decision set respectively.

[0028] In this step, the system breaks down medical knowledge. Specifically, the condition set consists of all medical concept entities related to the patient's state, serving as the input basis for reasoning; the decision set consists of medical concept entities related to clinical treatment or diagnostic conclusions, serving as the output target for reasoning. Traditional knowledge graphs often conflate all concepts, while this embodiment explicitly establishes a binary classification system of "condition-outcome" through symbolic representation.

[0029] Combination Figure 3 and Figure 4 As shown, taking the scenario of peri-implant disease as an example, "Suppuration on probing (SOP)," "Bleeding on probing (BOP)," "Probing depth (PD)," and "Bone loss (BL)" are conditional concepts that characterize the patient's objective status (stored in the condition set); while "Non-peri-implant health," "peri-implant health," and "Soft tissue surgery (CAF / CTG)" are decision-making concepts (stored in the decision set).

[0030] S103: Define the value space of the medical concept entity; when the value space of the medical concept entity is represented as a continuous medical variable, the continuous medical variable is discretized according to the corresponding preset threshold and transformed into a decidable logical condition.

[0031] In order for the computer to perform accurate logical deductions, the value space of the medical concept entity includes a numerical value space, a Boolean value space, and an enumeration value space.

[0032] Among them, for Boolean types (such as...) Figure 4 The values ​​for "inflammation signs" are "none / present" and enumeration types (such as...). Figure 4In the context of "Bleeding on Probing (BOP)", the value can be set to "single bleeding point only / multiple bleeding points" and can be directly used as a basis for logical judgment. However, for continuous medical variables, since continuous values ​​cannot be directly used for precise matching of rule antecedents, discretization mapping is required.

[0033] The specific implementation process is as follows: when the value space of the medical concept entity is the numerical value space and is represented as the continuous medical variable, the continuous medical variable is compared with the preset threshold and mapped to a discrete logical value that satisfies or does not satisfy the decidable logical condition. For example... Figure 4 As shown, the guidelines set the threshold for bone loss (BL) at 2 mm. If the entered BL value is 1.5 mm, the system will automatically compare it with 2 mm and map it as "logical true (meets the condition <2 mm)", thereby converting the medical indicator into a Boolean logic feature that can be strictly judged by the machine.

[0034] S104: Based on preset logical constraints, combine multiple condition concepts in the condition set to construct production reasoning rules, and represent the production reasoning rules as hyperedges.

[0035] This step is central to overcoming the "black box" problem of traditional medical models. The production reasoning rules adopt the "IF-THEN" rule form.

[0036] The antecedent of the rule is composed of at least one combination of the multiple conditional concepts based on logical conjunction (AND), logical disjunction (OR), or logical negation (NOT), and the consequent of the rule is composed of one or more of the decision concepts; the preset logical constraints include logical conjunction, logical disjunction, logical negation, and implication.

[0037] In computer graph theory, a regular edge can only connect two nodes. However, in complex medical decision-making, multiple symptoms often need to be "met simultaneously" for a diagnosis. Therefore, this embodiment introduces a "hyperedge" data structure, such as... Figure 2 As shown, the condition nodes on the left are wrapped in a set, such as rules. This indicates that condition node 1 and condition node 2 have a logical conjunction relationship (both are satisfied), jointly triggering decision node 1. A hyperedge allows multiple nodes to be enclosed and connected simultaneously, perfectly representing the complex nested logic of "multiple preconditions jointly triggering one or more postconditions".

[0038] S105: Take all the condition concepts in the condition set as condition nodes and all the decision concepts in the decision set as decision nodes. The condition nodes and the decision nodes together are used as hypergraph nodes. Combine the hyperedges with the hypergraph nodes to construct a hypergraph decision structure.

[0039] Specifically, each of the production reasoning rules corresponds to a hyperedge, which connects multiple condition nodes and points to at least one decision node. For example... Figure 3 As shown, this is an overview of the complete hypergraph decision structure constructed for peri-implant diseases. Conditional nodes are like the input layer of a neural network, decision nodes are like the output layer, and hyperedges are explicit, medically-compliant transmission channels.

[0040] To further enhance decision-making traceability, the hypergraph decision structure also includes global system information for storing the meaning of medical concepts, the source of thresholds, and literature references. This means that a hyperedge is not just a computational connection; its underlying layer also carries metadata indicating which year and which clinical guideline the rule originated from.

[0041] S106: Obtain the patient profile information at the current treatment time point, and match the patient profile information with the condition nodes in the hypergraph decision structure.

[0042] In practice, at least one of the patient's chief complaint, medical history, physical examination results, laboratory test results, and medical imaging information is obtained to construct the patient profile information. The system's input interface converts this multimodal information into the same value space format as S103, and performs numerical comparison and state binding with the various condition nodes at the lowest level of the hypergraph network.

[0043] S107: Based on the matching result between the patient profile information and the condition node, determine the activated hyperedge to filter out the activated production reasoning rules, generate the corresponding candidate decision set according to the decision node pointed to by the activated production reasoning rules, and output the candidate decision set and its corresponding reasoning path to achieve interpretable presentation of the clinical decision-making process.

[0044] The underlying computer algorithm traverses the hypergraph network: when all condition nodes enclosed by a certain hyperedge are lit up by the patient profile state, the state of that hyperedge changes from 0 to 1 (activated). At this time, the candidate decision set is jointly composed of decision concepts pointed to by multiple activated production reasoning rules (that is, when multiple rules are satisfied at the same time, the system will comprehensively generate a set containing multiple treatment opinions).

[0045] The most prominent creative advantage of this embodiment lies in its strong compatibility: the hypergraph decision structure supports parallel modeling of medical knowledge from different sources under the same structured representation system, and dynamically activates corresponding rule combinations in the production reasoning rules corresponding to the medical knowledge from different sources. For example... Figure 5 As shown, Guideline 1 sets a probing depth (PD) for the diagnosis of peri-implantitis. A 6mm result is considered positive, while Guideline 2 sets... A PD value of 4mm is considered an early abnormality. The hypergraph structure allows these two rules, which differ in threshold, to exist in parallel within the system as two independent hyperedges. When the PD value input in the patient profile is 5mm, the system matching finds that the conjunctive condition of the hyperedge in Guideline 1 is not met, but the corresponding hyperedge in Guideline 2 is successfully activated, thereby generating candidate decisions based on Guideline 2.

[0046] Ultimately, the front-end UI will output the clinical decision results along with the corresponding combination of conditional concepts and reasoning rule paths, thus achieving an interpretable presentation of the clinical decision-making process. Combined with... Figure 6 The visualization of the decision-making process for the "treatment plan for peri-implant soft tissue defects" shows a complete chain of events clearly highlighted by the system: because the patient meets the criteria of "PSTD Class III" + "presence of inflammation" + "probing depth PD". "6mm" + "Presence of labial mucosal retraction" (i.e., the combination of the aforementioned conditions) Activate the corresponding connection rules Ultimately, the candidate decision was derived as "embedded healing / soft tissue augmentation." This achieved a fully transparent decision-making process that incorporates expert-level slow-thinking logic and is inspired by human cognition.

[0047] Through the above implementation methods, this invention unifies the representation of multi-source medical knowledge using a hypergraph structure and explicitly characterizes the logical relationships of multi-condition joint triggering decisions through production reasoning rules, enabling the clinical decision-making process to be presented in a structured and traceable manner. Compared with traditional decision-making methods based on experience or black-box models, this invention can clearly demonstrate the role of different medical factors in decision-making and their logical basis, helping to reduce the cognitive load in complex clinical scenarios and improve the consistency and reliability of clinical decisions.

[0048] Furthermore, the hypergraph clinical decision framework described in this invention supports parallel modeling of different clinical guidelines and expert rules under the same structured representation system, and can dynamically activate corresponding rule combinations according to patient profiles. This allows for the flexible generation of individualized clinical decision paths while maintaining the standardization of evidence-based medicine, demonstrating good scalability and practical application value.

[0049] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed and specific, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.

[0050] This background section is provided to generally present the context of the invention. The work of the currently named inventors, the work to the extent described in this background section, and aspects of this section that did not constitute prior art at the time of application are neither expressly nor impliedly acknowledged as prior art to the invention.

Claims

1. A method for constructing a hypergraph-based clinical decision-making framework inspired by human cognition, characterized in that, include: Based on the clinical application scenario of the target disease, determine the type of clinical decision-making task and collect medical knowledge related to the clinical decision-making task; Extract medical concept entities related to the clinical decision-making task from the medical knowledge. The medical concept entities include symptoms and signs, examination indicators, disease status and treatment plan. Divide the medical concept entities into conditional concepts that characterize the patient status and decisional concepts that characterize the diagnosis or treatment, so as to form conditional set and decisional set respectively. Define the value space of the medical concept entity; when the value space of the medical concept entity is represented as a continuous medical variable, the continuous medical variable is discretized according to the corresponding preset threshold and transformed into a decidable logical condition; Based on preset logical constraints, multiple conditional concepts in the condition set are combined to construct production reasoning rules, and the production reasoning rules are represented as hyperedges. All conditional concepts in the condition set are used as conditional nodes, and all decisional concepts in the decision set are used as decisional nodes. The conditional nodes and the decisional nodes are used together as hypergraph nodes. The hyperedges are combined with the hypergraph nodes to construct a hypergraph decision structure. Obtain patient profile information at the current treatment time point, and match the patient profile information with the condition nodes in the hypergraph decision structure; Based on the matching results between the patient profile information and the condition nodes, the activated hyperedges are determined to filter out the activated production reasoning rules. The corresponding candidate decision set is generated according to the decision nodes pointed to by the activated production reasoning rules, and the candidate decision set and its corresponding reasoning path are output to achieve an interpretable presentation of the clinical decision-making process.

2. The method for constructing a hypergraph-based clinical decision-making framework inspired by human cognition according to claim 1, characterized in that, The clinical decision-making task types include at least one of diagnostic decision-making, treatment decision-making, decision-making to obtain further information, and decision-making to delay intervention.

3. The method for constructing a hypergraph-based clinical decision-making framework inspired by human cognition according to claim 1, characterized in that, The medical knowledge mentioned comes from at least one of the following: clinical guidelines, expert consensus, evidence-based medicine literature, and clinical practice experience.

4. The method for constructing a hypergraph-based clinical decision-making framework inspired by human cognition according to claim 1, characterized in that, The value space of the medical concept entity includes a numerical value space, a Boolean value space, and an enumerated value space.

5. The method for constructing a hypergraph-based clinical decision-making framework inspired by human cognition according to claim 4, characterized in that, When the value space of the medical concept entity is represented as a continuous medical variable, the continuous medical variable is discretized according to the corresponding preset threshold and transformed into a decidable logical condition, including: When the value space of the medical concept entity is the numerical value space and is represented as the continuous medical variable, the continuous medical variable is compared with the preset threshold and mapped to a discrete logical value that satisfies or does not satisfy the determinable logical condition.

6. The method for constructing a hypergraph-based clinical decision-making framework inspired by human cognition according to claim 1, characterized in that, The production reasoning rules adopt the "if-then" rule form; The antecedent of the rule is composed of at least one combination of the multiple conditional concepts based on logical conjunction, logical disjunction, or logical negation, and the consequent of the rule is composed of one or more of the decision concepts; the preset logical constraints include logical conjunction, logical disjunction, logical negation, and implication.

7. The method for constructing a hypergraph-based clinical decision-making framework inspired by human cognition according to claim 1, characterized in that, Each of the production reasoning rules corresponds to a superedge, which connects multiple condition nodes and points to at least one decision node.

8. The method for constructing a hypergraph-based clinical decision-making framework inspired by human cognition according to claim 1, characterized in that, The hypergraph decision structure also includes global system information for storing the meaning of medical concepts, the source of thresholds, and literature references.

9. The method for constructing a hypergraph-based clinical decision-making framework inspired by human cognition according to claim 1, characterized in that, The process of obtaining patient profile information at the current treatment time includes: The patient profile information is constructed by acquiring at least one of the patient's chief complaint information, consultation record, physical examination results, laboratory test results, and medical imaging information.

10. The method for constructing a hypergraph-based clinical decision-making framework inspired by human cognition according to claim 1, characterized in that, The hypergraph decision structure supports parallel modeling of medical knowledge from different sources under the same structured representation system; The process of selecting activated production reasoning rules includes: dynamically activating corresponding rule combinations from production reasoning rules corresponding to medical knowledge from different sources, based on the patient profile information.

Images