Diagnostic factor construction method, disease recommendation method and device based on diagnostic factor, and computer readable medium

Abstract

The invention provides a construction method of a diagnostic factor. The construction method comprises the following steps: a data acquisition and standardization step; obtaining multi-source heterogeneous clinical data of the subject; the clinical data comprises unstructured text data and structured data; performing standardization processing on the clinical data; an evidence aggregation and factor generation step; performing entity recognition and semantic analysis on the unstructured text data by adopting a natural language processing technology, extracting clinical term entities, and mapping the clinical term entities to a standard medical term system to form standardized clinical term entities; the structured data and the standardized clinical term entities form clinical evidences; and collecting the clinical evidence of the subject to the corresponding diagnostic factor based on a predefined diagnostic factor template library to obtain the diagnostic factor of the subject.

Description

Technical Field

[0001] This application belongs to the field of medical artificial intelligence and clinical decision support technology, specifically involving diagnostic factor construction methods, disease recommendation methods based on diagnostic factors, devices, and computer-readable media. Background Technology

[0002] With the deepening application of computer technology, big data analytics, and artificial intelligence in the medical field, clinical decision support systems have shown great potential in improving diagnostic and treatment efficiency and reducing misdiagnosis and missed diagnosis. Especially in the modern patient-centered medical model, how to achieve efficient and accurate disease identification and recommendation based on patients' multi-source heterogeneous clinical data has become a key direction in the research and development of smart medical systems.

[0003] Currently, commonly used auxiliary diagnostic methods in clinical practice mainly rely on physicians' experience or simple rule-matching techniques to compare patients' chief complaints, present medical history, and basic physical signs upon admission with known diseases to provide preliminary diagnostic suggestions. Some information systems have attempted to introduce machine learning models, using structured fields in electronic medical records (such as blood routine tests and liver function tests) for classification and prediction. However, these methods generally suffer from the following technical shortcomings.

[0004] 1. Insufficient utilization of clinical data and lack of a unified semantic expression structure; existing technologies mostly use raw clinical data directly for disease matching or classification, without effectively integrating and abstracting multi-source heterogeneous data, resulting in information redundancy, feature sparsity, semantic loss, and difficulty in comprehensively depicting the patient's condition.

[0005] 2. The diagnostic reasoning process lacks an intermediate logical layer and the reasoning mechanism is crude. Existing methods usually adopt a direct mapping model of "symptom → disease", which lacks interpretable intermediate reasoning units and cannot reflect the clinical doctor's thinking process "from evidence to judgment", resulting in poor interpretability of recommendation results and low doctor trust.

[0006] 3. The diagnostic dimensions are too narrow, neglecting the synergistic effect of multi-level and dynamic clinical features and patient attributes; current systems mostly rely on static chief complaints and do not fully integrate multi-dimensional information such as the patient's own characteristics, clinical history, trend of changes in signs, and dynamic evolution of examination and test results, making it difficult to deal with complex, atypical or comorbid situations.

[0007] 4. Lack of structured and scalable disease reasoning mechanisms; existing technologies are mostly based on rules or black box models, which make it difficult to support continuous optimization of diagnostic logic and knowledge updates, thus limiting the clinical adaptability and intelligence level of the system. Summary of the Invention

[0008] In view of the above problems, this application aims to propose a diagnostic factor construction method, a disease recommendation method and device based on diagnostic factors, and a computer-readable medium, which effectively solves the problems of data fragmentation, coarse reasoning, and poor interpretability in the prior art.

[0009] The method for constructing diagnostic factors in this application includes the following steps: Data collection and standardization steps; acquiring heterogeneous clinical data from multiple sources of subjects; this clinical data includes unstructured text data and structured data; standardizing the clinical data; The evidence aggregation and factor generation steps involve: using natural language processing technology to perform entity recognition and semantic parsing on unstructured text data, extracting clinical terminology entities, and mapping them to a standard medical terminology system to form standardized clinical terminology entities; structured data and standardized clinical terminology entities constitute clinical evidence; and the subjects' clinical evidence is aggregated into corresponding diagnostic factors based on a predefined diagnostic factor template library to obtain the subjects' diagnostic factors.

[0010] Preferably, the diagnostic factors are temporal; past diseases and past symptoms are integrated into the diagnostic factors; and clinical data from past and current medical visits are integrated into the diagnostic factors.

[0011] Preferably, the diagnostic factor is updatable; the diagnostic factor is dynamically updated as the subject's diagnosis and treatment progresses, and the update of the multidimensional diagnostic factor is triggered by new examination results, test results, or changes in symptoms.

[0012] Preferably, the diagnostic factors play different roles for different diseases, and are graded according to their importance to the disease, with different weights assigned to different grades.

[0013] The present application discloses a disease recommendation method based on diagnostic factors, wherein the diagnostic factors are those constructed by the aforementioned diagnostic factor construction method; the disease recommendation method includes: Construct a knowledge graph linking diagnostic factors and diseases; this knowledge graph is a many-to-many association network with diseases as nodes and diagnostic factors as edges; each edge is assigned a level, indicating the strength of diagnostic support for the disease by the diagnostic factor; Comprehensive recommendation weight calculation; for all multidimensional diagnostic factors of the subject, retrieve the set of candidate diseases associated with them; for each disease in the set of candidate diseases, calculate the comprehensive recommendation weight of the disease based on its associated diagnostic factors; Output interpretable recommendation results; form a disease ranking list based on comprehensive recommendation weights, with each item including: a list of supporting factors, clues to missing key evidence, and a visualized reasoning path.

[0014] Preferably, the basic weights are determined; the basic weights of the diagnostic factors are determined according to their preset levels. Weight adjustment; adjust the weights based on the attributes of the diagnostic factors, building upon the base weights. Multi-factor synergy; weighting adjustments are made considering the combined effects of multiple diagnostic factors; Exclusion and identification; reduce the weight of mismatched diseases through negative exclusion; increase the weight of matching diseases through clinical pathway temporal constraints.

[0015] The disease recommendation device based on diagnostic factors in this application includes: a data acquisition and standardization unit, a diagnostic factor construction unit, a disease reasoning engine unit, and a result generation unit; The data acquisition and standardization unit is used to obtain patients' multi-source heterogeneous clinical data from the hospital's information system and perform preliminary cleaning and standardization processing. The diagnostic factor building unit transforms standardized clinical data into medically meaningful diagnostic factors. The disease reasoning engine unit performs disease reasoning based on the diagnostic factors generated by the diagnostic factor construction module unit; The results generation unit presents the disease reasoning results from the disease reasoning engine unit in an intuitive and interpretable form.

[0016] Preferably, the diagnostic factor construction unit uses natural language processing technology to perform entity recognition and semantic parsing on unstructured text data in multi-source heterogeneous clinical data, extracts clinical terminology entities, and maps them to a standard medical terminology system to form standardized clinical terminology entities; structured data and standardized clinical terminology entities constitute clinical evidence; the clinical evidence of the subjects is collected into the corresponding diagnostic factors based on a predefined diagnostic factor template library to obtain the diagnostic factors of the subjects.

[0017] Preferably, the disease reasoning engine unit constructs a knowledge graph relating diagnostic factors to diseases; this knowledge graph is a many-to-many association network with diseases as nodes and diagnostic factors as edges; each edge is assigned a level, indicating the strength of diagnostic support for the disease by the diagnostic factor; the disease reasoning engine unit calculates a comprehensive recommendation weight; for all diagnostic factors of the subject, it retrieves the set of candidate diseases associated with them; for each disease in the set of candidate diseases, it calculates the comprehensive recommendation weight of the disease based on its associated diagnostic factors, which is used as the disease reasoning result. The results generation unit uses a comprehensive recommendation weight to form a disease ranking list, each item including: a list of suspected diseases and a visualized reasoning path.

[0018] The computer-readable medium of this application stores a computer program that, when executed, performs the aforementioned disease recommendation method based on diagnostic factors.

[0019] This application proposes diagnostic factors as core semantic units connecting raw clinical data with the final disease diagnosis. These diagnostic factors are not single indicators, but rather a multidimensional, structured clinical feature representation system. Their construction encompasses multi-source information including textual information from patient admission records (such as chief complaint, present illness, and past medical history), physical examination results, auxiliary examination results, laboratory test data, and their dynamic trends. Through natural language processing, data standardization, and feature fusion technologies, dispersed clinical observations are transformed into high-order feature units with medical semantics. This application also proposes a technical path for disease reasoning based on diagnostic factors. By constructing a knowledge graph linking diagnostic factors and diseases, and combining it with rule-based reasoning, an intelligent mapping from "multidimensional diagnostic factor combinations" to "potential disease sets" is achieved. This process not only supports rapid matching of typical cases but also identifies potential diseases, rare diseases, and critical illnesses with atypical presentations, improving diagnostic coverage and robustness. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the disease recommendation method and device based on diagnostic factors according to this application.

[0021] Figure 2 This is a schematic diagram illustrating the NLP processing of unstructured text data in this application.

[0022] Figure 3 Example of a diagnostic factor library associated with cerebral infarction.

[0023] Figure 4 Example of a diagnostic factor library associated with subcleft palate.

[0024] Figure 5 This is a diagram illustrating the symptoms of single and double symptom formation. Detailed Implementation

[0025] The disease recommendation method and apparatus based on diagnostic factors of this application comprises the following two core components: a multidimensional diagnostic factor construction mechanism and a disease reasoning mechanism based on diagnostic factors. These are described in detail below. 1. Multidimensional Diagnostic Factor Construction Mechanism - A Semantic Bridge for Clinical Intelligent Diagnosis This application proposes the innovative concept of "diagnostic factors" for the first time, representing a paradigm shift in the field of clinical decision support. This theory achieves intelligent mapping from raw medical information to clinical diagnosis by constructing a three-layer cognitive framework of "clinical data - diagnostic factors - disease entities". As an intermediate semantic unit connecting raw clinical data and disease diagnosis, a diagnostic factor is not a single indicator, but a structured feature unit with clear medical meaning, fused from multiple related clinical indicators such as symptoms, physical examination, signs, past illnesses, current illnesses, examinations, tests, and examination and test results. It transforms raw clinical data into comprehensive evidence used to characterize a specific pathological state or clinical manifestation, converting it into standard medical terminology. Diagnostic factors have the following characteristics: 1) Multidimensionality: Each diagnostic factor integrates information from different data sources, including but not limited to: textual information in admission records (such as chief complaint, present illness, past medical history); physical examination results (such as lung rales, abdominal tenderness); laboratory test results (such as elevated white blood cell count, abnormal C-reactive protein); imaging examination conclusions (such as "exudative shadow in the lower lobe of the right lung"); vital sign trends (such as persistent high fever for more than 3 days); treatment response information (such as a decrease in body temperature after antibiotic use).

[0026] 2) Temporality: Previous medical history, such as past diseases and symptoms, is also incorporated into the diagnostic factors; clinical data from past and current medical visits jointly support the diagnostic factor system.

[0027] 3) Updatable: Diagnostic factors are dynamically updated as the patient’s diagnosis and treatment progresses. New test results, laboratory results, or changes in symptoms can trigger a reassessment of the factor status.

[0028] 4) Grading: Based on the importance of the diagnostic factors to the disease, a grading system is set and weights are assigned.

[0029] 5) Other attributes: Combine patient visit type, hospitalization, outpatient, emergency, etc., to set whether the diagnostic factor is recalled for outpatient or inpatient visits; if a disease depends on a certain diagnostic factor, it can be set as a necessary condition; at the same time, if the appearance of a certain diagnostic factor suggests that the recommendation for this disease is no longer appropriate, it can be set as a negative condition.

[0030] The process of constructing diagnostic factors includes the following steps: 1) Data collection and standardization; Obtain multi-source patient data from Hospital Information System (HIS), Electronic Medical Record System (EMR), Laboratory Information System (LIS), and Picture Archiving and Communication System (PACS); Natural Language Processing (NLP) technology is used to perform entity recognition and semantic parsing on unstructured text (such as hospital admission), extract key clinical terms (such as "coughing up yellow phlegm" and "chest pain"), and map them to the standard medical terminology system.

[0031] 2) Evidence aggregation and factor generation; Based on a predefined library of diagnostic factor templates (built by medical experts and knowledge engineers), matching clinical evidence is grouped under the corresponding diagnostic factor.

[0032] Example: When the system recognizes two pieces of evidence, "cough" and "examination results: pulmonary infiltrate", it automatically activates and generates the diagnostic factor "pneumonia".

[0033] 2. Disease inference mechanisms based on diagnostic factors Based on the construction of multidimensional diagnostic factors, this application further proposes a mapping reasoning method from diagnostic factors to diseases, realizing intelligent recommendation from "evidence combination" to "potential disease".

[0034] 1) The reasoning mechanism includes the following steps; Clinical data layer → Diagnostic factor layer → Disease hypothesis layer → Decision optimization layer.

[0035] 2) Construct a knowledge graph of the association between diagnostic factors and diseases; Establish a many-to-many relationship network with "disease" as nodes and "diagnostic factors" as edges; Each edge is accompanied by a rating, which indicates the strength of the diagnostic factor’s support for the diagnosis of the corresponding disease (which can be obtained based on literature statistics, clinical guidelines or real-world data mining). It supports hierarchical structures, such as "pneumonia" being divided into subcategories like "bacterial pneumonia" and "viral pneumonia," corresponding to different combinations of factors.

[0036] 3) Multi-factor collaborative reasoning engine; For all diagnostic factors activated by the current patient, retrieve the set of candidate diseases associated with them; 4) Exclusion and identification mechanisms; Negative diagnostic factors (such as "no hemoptysis" to exclude pulmonary tuberculosis, and "normal CRP" to reduce the likelihood of bacterial infection) are introduced to reduce scores for mismatched diseases. Support time constraints based on clinical pathways (e.g., "fever before cough" is more supportive of respiratory infections).

[0037] 5) Output interpretable recommendation results; The recommendations are presented as a "ranked list of diseases," with each item containing: A list of suspected diseases is displayed; the reasoning path is visualized (the factor-disease association path is displayed by highlighting the list of diagnostic factors).

[0038] 1. Does the chief complaint meet the criteria for tumor, cancer, chemotherapy, radiotherapy, leukemia, etc.? If it does, do not exclude suspected diseases. II. Does the chief complaint meet the criteria for hospitalization for XX treatment, such as requesting XX surgery or XX treatment? "Request for hospitalization" is recommended. III. The logic of odd / even symptoms; If a patient has only one or two diagnostic factors, the system will match whether the patient meets the single or double symptom logic for disease. The system will also consider the patient's age and gender. If the criteria are met, a recommended disease will be made. IV. Diagnostic Factor Logic; The logic for single / double symptoms is recommended to be empty, and the logic for the chief complaint is not satisfied. 1. Obtain the patient's symptoms (chief complaint, present illness), signs (physical examination), examinations, tests, examination and test results (covering physical examination and doctor's orders), and past medical history (past illnesses). After structuring the unstructured original text, entities are obtained, normalized, and hypernyms are identified; the weight of hypernyms is reduced. Past medical history; Inspection name; test name; descriptive name of the inspection result; Increase the weight of diagnostic factors within the scope of the chief complaint; 2. Based on the patient's diagnostic factors, match the disease and disease-related diagnostic factor information (disease name, diagnostic factor name, diagnostic factor level, whether it is a negative condition, whether it is a necessary condition, whether outpatient / inpatient recall is required; disease-related departments, disease-related age, high-incidence age, etc.). 3. Diagnostic factors are used to calculate the weight of diseases (diagnostic factor level + diagnostic attributes). The decision is based on a combination of factors, including the weights set by the diagnostic factor levels; weight calculations are performed for individual diagnostic factors and multiple diagnostic factors under different conditions; recommendations are not made if necessary conditions are not met, or if negative conditions are met; outpatient and inpatient cases are not recalled, and recommendations cannot be made individually; if another recall is available, joint recommendations are made based on this logic. 4. Rationalization of disease-level recommendations; Recommendations should be made based on the appropriate age group and gender for the disease; if there are common departments, high-incidence age groups, or high-incidence genders, their weights should be increased and the settings should be adjusted accordingly. 5. Sort (descending order); The diseases are sorted in descending order by weight, and are categorized as critical, acute, or infectious. The system supports operations such as comparing with hospital names, enabling the creation of new cases, prominent display, and differentiation between common and rare diseases.

[0039] The disease recommendation system based on diagnostic factors described in this application is illustrated in conjunction with the accompanying drawings. Figure 1 As shown, the system mainly consists of a data acquisition and standardization module, a diagnostic factor construction module, a disease inference engine module, and a result generation module.

[0040] 1. Data Acquisition and Standardization Module This module is responsible for acquiring multi-source heterogeneous patient data from the hospital's information system and performing preliminary cleaning and standardization processing to lay the foundation for subsequent semantic parsing.

[0041] Specifically, the system collects patients' initial clinical data in real time or at set intervals from hospital information systems (HIS), electronic medical record systems (EMR), laboratory information systems (LIS), picture archiving and communication systems (PACS), and equipment systems such as electrocardiogram and patient monitors, via API interfaces or data middleware. This initial clinical data includes, but is not limited to: Unstructured text data includes: patient's chief complaint, present illness, past medical history, personal history, family history, physical examination records, progress notes, surgical records, discharge summaries, etc. For the collected unstructured text data, the system calls its built-in Natural Language Processing (NLP) engine to accurately extract clinical terminology entities from the text.

[0042] Structured data includes: vital signs (temperature, pulse, respiration, blood pressure), laboratory test results (complete blood count, urinalysis, biochemistry, immunology, microbiology, etc.), imaging examination results (summaries of reports from CT, MRI, X-ray, ultrasound, etc.), and nursing records.

[0043] After acquiring the data, the module performs standardization processing on the data.

[0044] 2. Diagnostic Factor Construction Module This module is the core of this application, responsible for transforming standardized clinical data into multidimensional diagnostic factors with medical significance. Its workflow is as follows: The system maps identified clinical terminology entities to standard medical terminology systems, such as the International Classification of Diseases (ICD-11), the Systematic Clinical Terminology Collection (SNOMED CT), or localized medical terminology dictionaries. This step ensures the standardization and consistency of terminology and resolves the issues of synonyms and near-synonyms.

[0045] Then, the system aggregates evidence and generates factors based on a pre-built diagnostic factor template library. This template library is jointly maintained by medical experts and knowledge engineers. Each template defines the name of a diagnostic factor, the types of clinical evidence it contains, the weight, level, and attributes of each piece of evidence.

[0046] When the system identifies a match between a patient's clinical data and the evidence in the template, it activates and generates the corresponding diagnostic factors.

[0047] Furthermore, this module is also time-sensitive and updatable. The system integrates historical data from the patient's current and previous visits. For example, if the patient's medical history includes "type 2 diabetes," the system will generate a "type 2 diabetes (past history)" diagnostic factor. As the treatment progresses, new test results (such as a blood culture report of "pneumococcal pneumonia positive") will be incorporated, and the system will dynamically update the confidence level of the "highly suspected community-acquired pneumonia" factor, and may even refine it to "pneumococcal pneumonia."

[0048] 3. Disease Inference Engine Module This module performs disease reasoning based on the set of diagnostic factors generated by the diagnostic factor construction module. Its core is a pre-built diagnostic factor-disease association knowledge graph.

[0049] The reasoning process is as follows: 1) Candidate disease retrieval: The system retrieves the set of all candidate diseases associated with all activated diagnostic factors of the current patient.

[0050] 2) Comprehensive Weight Calculation: For each disease in the candidate disease set, the system calculates a comprehensive recommendation weight based on its associated diagnostic factors. The calculation process includes: A. Determining the basic weights: The basic weights of the diagnostic factors are determined based on their preset levels.

[0051] B. Weight Adjustment: Adjustments are made based on the attributes of the diagnostic factors.

[0052] C. Multifactor synergy: Consider the combined effect of multiple diagnostic factors.

[0053] D. Exclusion and Differentiation: The system introduces negative diagnostic factors for exclusion.

[0054] 4. Result Generation Module This module is responsible for presenting the calculation results of the disease inference engine to doctors in an intuitive and interpretable form.

[0055] The system sorts the candidate diseases in descending order based on their comprehensive recommendation weights, generating a final disease recommendation list. This list not only includes the ranking results but also provides rich auxiliary information, such as: Recommendation levels: Recommended in descending order of weight.

[0056] Supporting Factors List: Clearly lists all diagnostic factors that support the diagnosis of this disease, allowing doctors to clearly understand the basis for their recommendations.

[0057] Priority marking: Critically ill, acute, and infectious diseases are clearly marked to remind doctors to pay priority attention to them.

[0058] In summary, this application achieves deep integration and structured expression of multi-source clinical data by constructing an innovative intermediate semantic layer of "diagnostic factors," and on this basis, performs interpretable and intelligent disease reasoning. This effectively solves the problems of insufficient data utilization and uninterpretable reasoning processes in existing technologies, providing clinicians with powerful and reliable decision support.

Claims

1. A method for constructing a diagnostic factor, comprising the following steps: Data collection and standardization steps; acquiring heterogeneous clinical data from multiple sources of subjects; this clinical data includes unstructured text data and structured data; standardizing the clinical data; The evidence aggregation and factor generation steps involve: using natural language processing technology to perform entity recognition and semantic parsing on unstructured text data, extracting clinical terminology entities, and mapping them to a standard medical terminology system to form standardized clinical terminology entities; structured data and standardized clinical terminology entities constitute clinical evidence; and the subjects' clinical evidence is aggregated into corresponding diagnostic factors based on a predefined diagnostic factor template library to obtain the subjects' diagnostic factors.

2. The method for constructing diagnostic factors according to claim 1, characterized in that: The diagnostic factors are temporal; past diseases and symptoms are incorporated into the diagnostic factors; and clinical data from past and current medical visits are incorporated into the diagnostic factors.

3. The method for constructing diagnostic factors according to claim 1, characterized in that: The diagnostic factor is updatable; it is dynamically updated as the subject's treatment progresses, and new examination results, test results, or symptom changes trigger the update of the multidimensional diagnostic factor.

4. The method for constructing diagnostic factors according to claim 1, characterized in that: The diagnostic factors play different roles for different diseases, and are graded according to their importance to the disease, with different weights assigned to different grades.

5. A disease recommendation method based on diagnostic factors, wherein the diagnostic factors are diagnostic factors constructed by the diagnostic factor construction method according to any one of claims 1-4; the disease recommendation method includes: Construct a knowledge graph linking diagnostic factors and diseases; this knowledge graph is a many-to-many association network with diseases as nodes and diagnostic factors as edges. Each edge is accompanied by a rating, indicating the strength of the diagnostic factor's support for the diagnosis of the disease; Comprehensive recommendation weight calculation; for all multidimensional diagnostic factors of the subject, retrieve the set of candidate diseases associated with them; for each disease in the set of candidate diseases, calculate the comprehensive recommendation weight of the disease based on its associated diagnostic factors; Output interpretable recommendation results; form a disease ranking list based on comprehensive recommendation weights, with each item including: a list of supporting factors, clues to missing key evidence, and a visualized reasoning path.

6. The disease recommendation method based on diagnostic factors according to claim 5, characterized in that: The calculation of the overall recommendation weight includes: Basic weights are determined based on the preset levels of the diagnostic factors. Weight adjustment; adjust the weights based on the attributes of the diagnostic factors, building upon the base weights. Multi-factor synergy; weighting adjustments are made considering the combined effects of multiple diagnostic factors; Exclusion and identification; reduce the weight of mismatched diseases through negative exclusion; increase the weight of matching diseases through clinical pathway temporal constraints.

7. A disease recommendation device based on diagnostic factors, comprising: Data acquisition and standardization unit, diagnostic factor construction unit, disease reasoning engine unit, and result generation unit; The data acquisition and standardization unit is used to obtain patients' multi-source heterogeneous clinical data from the hospital's information system and perform preliminary cleaning and standardization processing. The diagnostic factor building unit transforms standardized clinical data into medically meaningful diagnostic factors. The disease reasoning engine unit performs disease reasoning based on the diagnostic factors generated by the diagnostic factor construction module unit; The results generation unit presents the disease reasoning results from the disease reasoning engine unit in an intuitive and interpretable form.

8. The disease recommendation device based on diagnostic factors according to claim 7, characterized in that: The diagnostic factor construction unit uses natural language processing technology to perform entity recognition and semantic parsing on unstructured text data from multi-source heterogeneous clinical data, extracts clinical terminology entities, and maps them to a standard medical terminology system to form standardized clinical terminology entities. The structured data and standardized clinical terminology entities constitute clinical evidence. The clinical evidence of the subjects is collected into the corresponding diagnostic factors based on a predefined diagnostic factor template library to obtain the diagnostic factors of the subjects.

9. The disease recommendation device based on diagnostic factors according to claim 8, characterized in that: The disease reasoning engine unit constructs a knowledge graph linking diagnostic factors and diseases. This knowledge graph is a many-to-many network with diseases as nodes and diagnostic factors as edges. Each edge is assigned a level, indicating the strength of diagnostic support for the disease. The disease reasoning engine unit calculates the comprehensive recommendation weight. For all diagnostic factors of the subject, it retrieves the set of candidate diseases associated with them. For each disease in the set of candidate diseases, it calculates the comprehensive recommendation weight of the disease based on its associated diagnostic factors, which is then used as the disease reasoning result. The results generation unit uses a comprehensive recommendation weight to form a disease ranking list, each item including: a list of suspected diseases and a visualized reasoning path.

10. A computer-readable medium having a computer program stored thereon, which, when executed, performs the disease recommendation method based on diagnostic factors as described in any one of claims 5-6.

Images