A knowledge graph organization system for gastroenterology

Abstract

The present invention relates to the technical field of knowledge graph organization, and specifically to a gastroenterology knowledge graph organization system, comprising: a knowledge graph construction module for generating a weighted knowledge graph; a diagnostic relevance evaluation module for calculating and generating a diagnostic relevance score for a set of diagnostic paths corresponding to a real-time topological vector; a decision feedback capture module for generating a decision feedback data set including a final adopted diagnosis, a user operation score, and a decision deviation; and a weight iterative optimization module for generating updated evidence-based weights to achieve closed-loop self-optimization of the knowledge graph. The present invention achieves closed-loop self-optimization of the knowledge graph.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge graph organization, and in particular to a gastroenterology knowledge graph organization system. Background Art

[0002] In the field of clinical diagnosis, especially in specialties like gastroenterology that rely on complex differential diagnoses, traditional diagnostic methods primarily rely on physicians' personal clinical experience, static medical textbooks, and treatment guidelines. These methods are increasingly limited in the face of growing volumes of medical data and complex conditions. These limitations are primarily manifested in lagging knowledge updates, inefficient data integration, and an inability to dynamically optimize the diagnostic process. This situation makes it difficult for physicians, especially younger ones, to quickly and accurately form differential diagnoses when faced with atypical or complex cases, impacting the timeliness and accuracy of diagnoses.

[0003] The above-mentioned status quo and shortcomings are mainly due to the limitations of knowledge integration methods and clinical data processing technologies. First, medical knowledge exists in an isolated form in literature, guidelines, and doctors' experience, which leads to subjectivity and lag in the application of knowledge. Second, electronic medical records contain a large amount of valuable unstructured information. Traditional methods have difficulty in automatically and efficiently extracting this information and integrating it with structured data, resulting in a lack of clinical data dimensions. The end result is that when doctors face patients with complex conditions, they are unable to quickly and comprehensively weigh all diagnostic possibilities under a huge cognitive load. This may not only delay the diagnosis and treatment of critical conditions such as malignant diseases, but also increase the risk of missed diagnosis and misdiagnosis, and may lead to unnecessary examinations, increasing medical costs. Summary of the Invention

[0004] The purpose of the present invention is to provide a gastroenterology knowledge graph organization system to solve the problems existing in the background technology.

[0005] To solve the above technical problems, the present invention provides a gastroenterology knowledge graph organization system, comprising:

[0006] The knowledge graph construction module is used to extract medical concept entities and semantic relationships from medical data sources, and assign an initial evidence-based weight to each semantic relationship based on the preset evidence-based medicine quantification rules, thereby generating a weighted knowledge graph;

[0007] A diagnostic relevance evaluation module is configured to receive a real-time topological vector representing a patient's condition, determine a set of diagnostic pathways based on the weighted knowledge graph and the real-time topological vector in combination with an initial evidence-based weight, and calculate and generate a diagnostic relevance score corresponding to the set of diagnostic pathways;

[0008] The decision feedback capture module is used to record the user's interactive operations on the diagnosis relevance score, compile the recorded interactive operations into the user operation history, and generate a decision feedback dataset containing the final adopted diagnosis, user operation score and decision deviation based on the user operation history and user level information;

[0009] The weight iterative optimization module is used to modify the initial evidence-based weights on the corresponding diagnostic path in the weighted knowledge graph according to the decision feedback data set and the preset weight update rules, and generate updated evidence-based weights to achieve closed-loop self-optimization of the knowledge graph.

[0010] Preferably, the knowledge graph construction module is used to perform the following operations: each semantic relationship between medical concept entities is processed through preset evidence-based medicine quantification rules according to the evidence level, publication time and journal impact factor of its source to generate an initial evidence-based evidence weight, wherein the evidence-based medicine quantification rules are configured to assign a higher initial evidence-based evidence weight to semantic relationships with higher evidence level, newer publication time and higher journal impact factor.

[0011] Preferably, the system also includes a topological vector parsing module, and the real-time topological vector is generated by the topological vector parsing module. The topological vector parsing module is used to extract the topological coordinates representing the anatomical location of the lesion, the category code representing the pathological nature, the quantitative score representing the severity of the lesion, and the descriptor vector representing the morphological characteristics of the lesion from the patient's electronic medical record, and combine the above information to generate a real-time topological vector for processing by the diagnostic relevance evaluation module.

[0012] Preferably, when calculating the diagnostic relevance score, the diagnostic relevance assessment module is further configured to:

[0013] Obtaining a set of typical topological vectors representing historical cases;

[0014] Calculate the topological similarity between the real-time topology vector and each vector in the typical topology vector set;

[0015] The topological similarity was combined with the product of the initial evidence-based weights along the diagnostic pathway to generate a diagnostic relevance score.

[0016] Preferably, the system further comprises a user level assessment module, which is used to:

[0017] Generate user-level parameters based on the user's job level information and user operation history, where the user operation history includes operation accuracy and response time;

[0018] The diagnostic relevance evaluation module is further configured to receive user-level parameters and perform weighted adjustment on the diagnostic relevance score using the user-level parameters to generate a user-level-adjusted diagnostic relevance score.

[0019] Preferably, the system further comprises a diagnostic confusion assessment module, which is used to:

[0020] Normalize the diagnostic relevance scores corresponding to the diagnostic pathway set to generate a diagnostic posterior probability distribution;

[0021] Based on the diagnostic posterior probability distribution, the diagnostic confusion index used to characterize the current diagnostic uncertainty is calculated, and the diagnostic confusion index is output to the user interface for risk warning.

[0022] Preferably, when generating the decision feedback data set, the decision feedback capture module is further configured to:

[0023] The ranking of the finally adopted diagnosis in the diagnostic path set sorted by score is combined with the user operation score generated based on the user operation history to calculate the decision bias; wherein the decision bias is configured to amplify the ranking of the finally adopted diagnosis when the user operation score is low.

[0024] Preferably, the weight iterative optimization module is used to:

[0025] The decision bias is combined with the user operation score to generate a weight correction, where the size of the weight correction is positively correlated with both the decision bias and the user operation score.

[0026] The weight correction amount is then applied to the diagnostic pathway corresponding to the final adopted diagnosis to generate an updated evidence-based weight.

[0027] Preferably, the weight iterative optimization module is further used to:

[0028] Within a preset verification cycle, historical cases are re-diagnosed based on the updated evidence-based weights to generate new system accuracy rates;

[0029] The new system accuracy is compared with the system accuracy before the update. If and only if the improvement value of the new system accuracy exceeds a preset accuracy gain threshold, the updated evidence-based weight of evidence will be solidified and adopted in the subsequent diagnostic process.

[0030] Beneficial Effects Compared with the prior art, the present invention has the following beneficial effects:

[0031] The present invention provides a gastroenterology knowledge graph organization system, which has the following beneficial effects:

[0032] (1) Through the collaborative work of the knowledge graph construction module, the diagnostic relevance evaluation module, the decision feedback capture module, and the weight iterative optimization module, a complete closed loop of diagnostic decision support and knowledge evolution is constructed. Compared with the traditional diagnostic model that relies on the doctor's personal experience and static medical literature, this system transforms scattered knowledge and data into a dynamic, computable weighted knowledge graph, and can learn and optimize from each clinical practice, realizing the closed-loop self-optimization of the knowledge graph.

[0033] (2) Through the refined operations performed by its knowledge graph construction module, it can assign objective initial evidence-based weight to medical knowledge based on the preset quantitative rules of evidence-based medicine. At the same time, through the topological vector parsing module, the system can automatically extract and combine standardized real-time topological vectors from unstructured electronic medical records. This design significantly enhances the system's ability to process complex clinical data. It not only ensures the authority and timeliness of the knowledge source, but also enables the system to accurately capture the unique pathological characteristics of each patient.

[0034] (3) Through its diagnostic relevance assessment module, a historical case comparison mechanism is introduced, calculating the similarity between the current patient's real-time topological vector and the typical topological vector of historical cases, thereby integrating evidence-based medicine theory and clinical practice experience. Through the user-level assessment module and the diagnostic confusion assessment module, the system introduces user-level parameters to personalize the weighting of diagnostic scores and calculates the diagnostic confusion index for risk warning, significantly improving the diagnostic accuracy and safety of complex and difficult cases. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention, and those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0036] Figure 1 It is a logic block diagram of the system of the present invention. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0038] See also Figure 1The present invention provides a digestive medicine knowledge graph organization system, comprising:

[0039] The knowledge graph construction module is used to extract medical concept entities and semantic relationships from medical data sources, and assign an initial evidence-based weight to each semantic relationship based on the preset evidence-based medicine quantification rules, thereby generating a weighted knowledge graph;

[0040] A diagnostic relevance evaluation module is configured to receive a real-time topological vector representing a patient's condition, determine a set of diagnostic pathways based on the weighted knowledge graph and the real-time topological vector in combination with an initial evidence-based weight, and calculate and generate a diagnostic relevance score corresponding to the set of diagnostic pathways;

[0041] The decision feedback capture module is used to record the user's interactive operations on the diagnosis relevance score, compile the recorded interactive operations into the user operation history, and generate a decision feedback dataset containing the final adopted diagnosis, user operation score and decision deviation based on the user operation history and user level information;

[0042] The weight iterative optimization module is used to modify the initial evidence-based weights on the corresponding diagnostic path in the weighted knowledge graph according to the decision feedback data set and the preset weight update rules, and generate updated evidence-based weights to achieve closed-loop self-optimization of the knowledge graph.

[0043] This embodiment provides a knowledge graph organization system for gastroenterology. In a gastroenterology outpatient clinic setting, this system can play a key role in assisting diagnosis when a doctor encounters a patient complaining of upper abdominal pain. First, the knowledge graph construction module processes a vast amount of medical literature to construct a weighted knowledge graph that represents the relationship between gastric ulcers and upper abdominal pain. Next, after the doctor enters the patient's detailed medical condition, the system generates a real-time topological vector. The diagnostic relevance assessment module calculates relevance scores for potential diagnoses, such as gastric ulcers, on this graph. After consulting the score list and conducting further examinations, the doctor ultimately adopts the diagnosis of gastric ulcer. The decision feedback capture module records the doctor's actions and generates feedback data based on their rank information. Finally, the weight iterative optimization module, based on this feedback data, adjusts the weight of the diagnostic path from upper abdominal pain to gastric ulcers, thereby enhancing the system's accuracy in handling similar cases in the future. Through the collaborative work of these four modules, this system not only provides accurate reference for a single diagnosis but also forms a closed loop of continuous learning and self-optimization, continuously improving its auxiliary diagnosis capabilities in the field of gastroenterology.

[0044] Example 1

[0045] The knowledge graph construction module is used to perform the following operations: each semantic relationship between medical concept entities is processed through the preset evidence-based medicine quantification rules according to the evidence level, publication time and journal impact factor of its source to generate an initial evidence-based evidence weight. Among them, the evidence-based medicine quantification rules are configured to give higher initial evidence-based evidence weights to semantic relationships with higher evidence level, newer publication time and higher journal impact factor.

[0046] The system also includes a topological vector parsing module, and a real-time topological vector is generated by the topological vector parsing module. The topological vector parsing module is used to extract topological coordinates representing the anatomical location of the lesion, category codes representing the pathological properties, quantitative scores representing the severity of the lesion, and descriptor vectors representing the morphological characteristics of the lesion from the patient's electronic medical record, and combine the above information to generate a real-time topological vector for processing by the diagnostic relevance evaluation module.

[0047] The knowledge graph construction module generates initial evidence-based weights, which are effective in refining medical knowledge and quantifying its credibility. For example, given two pieces of diagnostic evidence pointing to gastric cancer, one from a recent Lancet publication (high-level evidence) and the other from a general medical textbook from ten years ago, the module calculates completely different weights using the following formula:

[0048] Normalization factor

[0049] ;

[0050] in, For the final evidence-based weight, the value range is normalized;

[0051] Code the categories for the level of evidence;

[0052] The year of publication, combined with (Current Year) constitutes the time decay factor;

[0053] is the impact factor of the source journal.

[0054] By this formula, the former is coded as having a higher level of evidence. , smaller year differences and higher impact factors , which received a significantly higher weight of evidence than the latter , ensuring the authority and timeliness of the knowledge graph.

[0055] At the same time, when facing a specific patient, the topological vector parsing module extracts key information from their electronic medical records, including: the topological coordinates of the anatomical position of the digestive tract encoded according to the standard anatomical model, the quantitative score of the severity of the lesion and the category ID of the pathological nature, and generates a real-time topological vector for accurate matching. :

[0056] ;

[0057] in, is the real-time topology vector;

[0058] The topological coordinates of the anatomical position of the digestive tract are encoded according to the standard anatomical model;

[0059] The category ID is the pathological property, such as inflammation, ulcer, and tumor;

[0060] It is a quantitative score for the severity of the lesion;

[0061] It is a descriptor vector of lesion morphology extracted from the endoscopy report based on NLP technology, such as irregular edges and dirty surface.

[0062] This process converts unstructured medical record text into standardized multidimensional vectors, providing an accurate and computable patient portrait for subsequent diagnostic relevance assessment, greatly improving the granularity and accuracy of diagnostic matching.

[0063] Example 2

[0064] When calculating the diagnostic relevance score, the diagnostic relevance assessment module is also used to:

[0065] Obtaining a set of typical topological vectors representing historical cases;

[0066] Calculate the topological similarity between the real-time topology vector and each vector in the typical topology vector set;

[0067] The topological similarity was combined with the product of the initial evidence-based weights along the diagnostic pathway to generate a diagnostic relevance score.

[0068] The diagnostic relevance evaluation module receives the patient real-time topology vector generated in Example 1 After that, it calculates the diagnostic relevance score The module combines the real-time topological vector with the knowledge base and the data related to a certain diagnosis. Typical topological vectors associated with (such as gastric signet ring cell carcinoma) For comparison, the topological similarity is calculated using the following formula:

[0069] ;

[0070] in, is the cosine similarity of topological vectors;

[0071] : real-time topology vector;

[0072] : typical topological vector;

[0073] k represents the dimension index of the topological vector;

[0074] It is a dimension valid indicator used to deal with the vector sparseness problem caused by missing information. When the feature is valid , otherwise 0.

[0075] This similarity score quantifies the degree of similarity between the current patient and a typical case of gastric signet ring cell carcinoma diagnosed in history in terms of lesion location, nature, severity and morphology. Ultimately, the module combines this similarity with the patient's symptom-to-diagnosis process. The weights of the path evidence chains are combined to generate the total score:

[0076]

[0077] in, For diagnosis Total relevance score;

[0078] is the patient's symptom combination;

[0079] It is the product of the evidence-based weights for each step in the diagnostic pathway.

[0080] The ith symptom in the symptom combination ( )

[0081] From the symptoms To diagnosis diagnostic pathways;

[0082] In this way, a case whose symptoms may be related to multiple diseases but whose endoscopic morphology is highly consistent with gastric signet ring cell carcinoma can be The scores will be significantly improved, which means that diagnostic recommendations are no longer simple symptom matching, but rather an in-depth and accurate assessment combined with practical experience of typical cases, which can effectively identify those atypical but clearly characterized difficult and complicated diseases.

[0083] Example 3

[0084] The system also includes a user level assessment module, which is used to:

[0085] Generate user-level parameters based on the user's job level information and user operation history, where the user operation history includes operation accuracy and response time;

[0086] The diagnostic relevance evaluation module is further configured to receive user-level parameters and perform weighted adjustment on the diagnostic relevance score using the user-level parameters to generate a user-level-adjusted diagnostic relevance score.

[0087] The system also includes a diagnostic confusion assessment module, which is used to:

[0088] Normalize the diagnostic relevance scores corresponding to the diagnostic pathway set to generate a diagnostic posterior probability distribution;

[0089] Based on the diagnostic posterior probability distribution, the diagnostic confusion index used to characterize the current diagnostic uncertainty is calculated, and the diagnostic confusion index is output to the user interface for risk warning.

[0090] The system introduces a user level assessment module to achieve personalized human-computer interaction and risk control. This module generates user level parameters for the current operating doctor through the following formula: :

[0091] ;

[0092] ;

[0093] in, is a user-level parameter;

[0094] Code the doctor's rank, for example: intern = 1, chief physician = 3;

[0095] Score operational history;

[0096] is the accuracy of historical operations;

[0097] is the response time;

[0098] is the weight coefficient;

[0099] n is the total number of historical operation records of the doctor.

[0100] An experienced chief physician with a history of fast and accurate operation, whose user level parameters The diagnostic relevance evaluation module will combine this parameter when generating the diagnostic score, for example, adding the item , which gives higher weight to the diagnostic pathways that senior physicians focus on, reflecting the respect and weighting for expert experience.

[0101] At the same time, in order to ensure the safety of diagnosis, the diagnosis confusion assessment module will conduct risk assessment on the score list. Converted into the corrected posterior probability of diagnosis , and then calculate the information entropy as the diagnostic confusion index :

[0102] ;

[0103] in, is the corrected posterior probability of diagnosis; is the total number of candidate diagnoses.

[0104] j is from 1 to The index variable ( Represents the total number of candidate diagnoses), used to traverse each candidate diagnosis .

[0105] represents the jth diagnosis in the candidate diagnosis set (e.g. For gastric ulcer, for gastritis, etc.), by The posterior probability is used to calculate the information entropy and obtain the confusion index that characterizes the uncertainty of diagnosis

[0106] When multiple candidate diagnoses When very close, If the value is high, the system interface will display a high-risk warning, indicating high diagnostic uncertainty and recommending a multidisciplinary consultation. This mechanism, by quantifying diagnostic ambiguity, provides a critical decision-making safety net for doctors. In complex scenarios where the system struggles to clearly distinguish between multiple possible diseases, it can provide effective early warnings and prevent missed or misdiagnoses.

[0107] Example 4

[0108] When generating a decision feedback dataset, the decision feedback capture module is also used to:

[0109] The ranking of the finally adopted diagnosis in the diagnostic path set sorted by score is combined with the user operation score generated based on the user operation history to calculate the decision bias; wherein the decision bias is configured to amplify the ranking of the finally adopted diagnosis when the user operation score is low.

[0110] The weight iterative optimization module is used to modify the initial evidence-based weights:

[0111] The decision bias is combined with the user operation score to generate a weight correction, where the size of the weight correction is positively correlated with both the decision bias and the user operation score.

[0112] The weight correction amount is then applied to the diagnostic pathway corresponding to the final adopted diagnosis to generate an updated evidence-based weight.

[0113] When a case of Crohn's disease is finally diagnosed clinically ( ), suppose that Crohn's disease is ranked third in the diagnostic ranking given by the system ( ). The decision feedback capture module will calculate the decision deviation at this time :

[0114] ;

[0115] in, is the decision bias factor;

[0116] systematic ranking for final confirmed diagnosis;

[0117] is the doctor's historical operation score;

[0118] is the deviation adjustment coefficient.

[0119] If the doctor is a young doctor with little experience, Lower, then The term will be larger, resulting in This design reflects the system's prudent learning strategy: for diagnoses selected by inexperienced doctors and ranked low, the system considers them to be very valuable learning signals, indicating that the system may have a large bias, so the bias value is increased.

[0120] Subsequently, the weight iterative optimization module corrects the knowledge path based on this deviation. It generates the weight correction amount and updates the weight using the following formula:

[0121] ;

[0122] in, is the updated weight;

[0123] is the basic learning rate; is the total number of candidate diagnoses, used for normalization;

[0124] As the indicator function, it ensures that only the weights on the final diagnosis path are corrected;

[0125] It is the chain of evidence for the final diagnosis.

[0126] The size of the weight correction and the decision deviation factor and the doctor's historical operation score This means that an experienced doctor ( High) found a large deviation in the system ( By applying this correction, the system increases the weight of all evidence along the diagnostic pathway for Crohn's disease. This allows for earlier and more accurate predictions of Crohn's disease in future patients with similar symptoms and pathological features, achieving precise and dynamic self-optimization of the knowledge graph.

[0127] Example 5

[0128] The weight iterative optimization module is also used to:

[0129] Within a preset verification cycle, historical cases are re-diagnosed based on the updated evidence-based weights to generate new system accuracy rates;

[0130] The new system accuracy is compared with the system accuracy before the update. If and only if the improvement value of the new system accuracy exceeds a preset accuracy gain threshold, the updated evidence-based weight of evidence will be solidified and adopted in the subsequent diagnostic process.

[0131] After a verification cycle of learning and accumulating multiple weight updates, the weight iteration optimization module does not immediately solidify the new weights permanently. It will start an iterative verification. The module will select a group of typical or difficult cases in history (a total of Example), use the update and after update The weight of the system is retrospectively re-diagnosed and the new system accuracy is calculated according to the following formula :

[0132] ;

[0133] in, Used to calculate the proportion of the final diagnosis ranking in the top 3;

[0134] It is the historical score of the doctor who performed the corresponding case at the time, which serves as a weighted item of cognitive conformity;

[0135] m represents the total number of historical cases used for retrospective rediagnosis

[0136] is the cognitive weight coefficient.

[0137] This accuracy rate not only takes into account the hard ranking, but also combines user perception data to measure whether the optimization is useful. Calculate the new system accuracy rate and system accuracy After that, the module will make the final decision:

[0138] ;

[0139] in, is the preset accuracy gain threshold.

[0140] Only when the accuracy improvement brought by the new weights is significant enough to exceed this threshold, the system will formally accept the learning results of this cycle and fully adopt them in subsequent diagnoses. On the contrary, if the improvement is not obvious or even decreases, the system will roll back to , to prevent invalid or harmful updates from polluting the knowledge base. This mechanism ensures that each iteration of the system is a true evolution, guaranteeing the robustness of the knowledge graph's self-optimization and continuous, high-quality performance improvement.

[0141] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A gastroenterology knowledge graph organization system, characterized by: include: The knowledge graph construction module is used to extract medical concept entities and semantic relationships from medical data sources, and assign an initial evidence-based weight to each semantic relationship based on the preset evidence-based medicine quantification rules, thereby generating a weighted knowledge graph; A diagnostic relevance evaluation module is configured to receive a real-time topological vector representing a patient's condition, determine a set of diagnostic pathways based on the weighted knowledge graph and the real-time topological vector in combination with an initial evidence-based weight, and calculate and generate a diagnostic relevance score corresponding to the set of diagnostic pathways; The decision feedback capture module is used to record the user's interactive operations on the diagnosis relevance score, compile the recorded interactive operations into the user operation history, and generate a decision feedback dataset containing the final adopted diagnosis, user operation score and decision deviation based on the user operation history and user level information; The weight iterative optimization module is used to modify the initial evidence-based weights on the corresponding diagnostic paths in the weighted knowledge graph according to the decision feedback dataset and the preset weight update rules, and generate updated evidence-based weights to achieve closed-loop self-optimization of the knowledge graph; The knowledge graph construction module is used to perform the following operations: each semantic relationship between medical concept entities is processed through the preset evidence-based medicine quantification rules according to the evidence level, publication time and journal impact factor of its source to generate an initial evidence-based evidence weight. Among them, the evidence-based medicine quantification rules are configured to give higher initial evidence-based evidence weights to semantic relationships with higher evidence level, newer publication time and higher journal impact factor.

2. A gastroenterology knowledge graph arrangement system according to claim 1, characterized in that: The system also includes a topological vector parsing module, and a real-time topological vector is generated by the topological vector parsing module. The topological vector parsing module is used to extract topological coordinates representing the anatomical location of the lesion, category codes representing the pathological properties, quantitative scores representing the severity of the lesion, and descriptor vectors representing the morphological characteristics of the lesion from the patient's electronic medical record, and combine the above information to generate a real-time topological vector for processing by the diagnostic relevance evaluation module.

3. A gastroenterology knowledge graph arrangement system according to claim 2, characterized in that: When calculating the diagnostic relevance score, the diagnostic relevance assessment module is also used to: Obtaining a set of typical topological vectors representing historical cases; Calculate the topological similarity between the real-time topology vector and each vector in the typical topology vector set; The topological similarity was combined with the product of the initial evidence-based weights along the diagnostic pathway to generate a diagnostic relevance score.

4. A gastroenterology knowledge graph arrangement system according to claim 3, characterized in that: Also included is a User Level Assessment module that: Generate user-level parameters based on the user's job level information and user operation history, where the user operation history includes operation accuracy and response time; The diagnostic relevance evaluation module is further configured to receive user-level parameters and perform weighted adjustment on the diagnostic relevance score using the user-level parameters to generate a user-level-adjusted diagnostic relevance score.

5. A gastroenterology knowledge graph arrangement system according to claim 4, characterized in that: It also includes a diagnostic confusion assessment module, which is used to: Normalize the diagnostic relevance scores corresponding to the diagnostic pathway set to generate a diagnostic posterior probability distribution; Based on the diagnostic posterior probability distribution, the diagnostic confusion index used to characterize the current diagnostic uncertainty is calculated, and the diagnostic confusion index is output to the user interface for risk prompting.

6. A gastroenterology knowledge graph arrangement system according to claim 5, characterized in that: When generating a decision feedback dataset, the decision feedback capture module is also used to: The ranking of the finally adopted diagnosis in the diagnostic path set sorted by score is combined with the user operation score generated based on the user operation history to calculate the decision bias; wherein the decision bias is configured to amplify the ranking of the finally adopted diagnosis when the user operation score is low.

7. A gastroenterology knowledge graph arrangement system according to claim 6, characterized in that: The weight iterative optimization module is used to modify the initial evidence-based weights: The decision bias is combined with the user operation score to generate a weight correction, where the size of the weight correction is positively correlated with both the decision bias and the user operation score. The weight correction amount is applied to the diagnostic pathway corresponding to the final adopted diagnosis to generate an updated evidence-based weight.

8. A gastroenterology knowledge graph arrangement system according to claim 7, characterized in that: The weight iterative optimization module is also used to: Within a preset verification cycle, historical cases are re-diagnosed based on the updated evidence-based weights to generate new system accuracy rates; The new system accuracy is compared with the system accuracy before the update. If and only if the improvement value of the new system accuracy exceeds a preset accuracy gain threshold, the updated evidence-based weight of evidence will be solidified and adopted in the subsequent diagnostic process.

Images