A pathological critical value warning method and system based on pathological knowledge graph

Abstract

The present invention discloses a pathology critical value early warning method and system based on pathology knowledge graph, which relates to the field of computer technology. The pathology critical value early warning method based on pathology knowledge graph includes the following steps: obtaining test index characteristic data; identifying key characteristic data; evaluating the test index risk level through an early warning model; judging the key factors affecting the test index risk level, and giving early warning reminders based on the key factors. The present invention evaluates the test index risk level through an early warning model, which helps to discover potential risks early, analyzes the test index risk level through a grade analysis algorithm, judges the key factors affecting the risk level, and gives early warning reminders, which can achieve comprehensive analysis and timely response to risks, thereby improving the efficiency and effectiveness of pathology critical value management.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a pathology critical value early warning method and system based on a pathology knowledge graph. Background Art

[0002] Clinical critical values ​​are often seen in tests, pathology and other examinations. In traditional laboratories, most critical values ​​are numerical and can be set by simple rules. Pathological diagnosis is an important technical means in clinical medical testing. It is mainly used for small biopsy pathological specimens or surgically removed pathological tissue specimens through gastroenteroscopy, hysteroscopy, colposcopy, etc., and finally obtains a pathological diagnosis report through microscopic observation and testing of related items such as molecular pathology.

[0003] The critical values ​​in the pathology report, also known as the pathological critical values, are different from the traditional critical values ​​for testing. Most of them are pathological diagnosis reports with a text report structure. For example, the clinical diagnosis is epidermoid cyst of the skin. After pathological biopsy, the final pathological diagnosis is malignant melanoma. Clinicians have completely different clinical treatment principles for "epidermoid cyst" and "malignant melanoma". "Epidermoid cyst" can be cured by surgical resection, while "malignant melanoma" requires expanding the tumor resection boundary and ensuring that there is no tumor residue at the surgical edge. Other systemic treatments such as chemotherapy and immunotherapy are required in the future. Patients need regular testing and follow-up to prevent metastasis to other parts. At present, the critical values ​​in pathology reports are relatively complex in definition, scope and classification. It is difficult to achieve the final timely clinical feedback effect by simply setting numerical standards.

[0004] The existing technology is not convenient for identifying key feature data, and cannot effectively screen out the data that has the greatest impact on risk assessment. It is also not convenient for building early warning models, and cannot evaluate the risk level of test indicators through early warning models. It is not convenient for early detection of potential risks, and it is not convenient for converting the risk trend of test indicators at future moments into actual risk assessments, which reduces the efficiency of early warning. At the same time, it is not convenient for analyzing the risk level of test indicators, determining the key factors that affect the risk level, and issuing early warning reminders, which makes it impossible to achieve a comprehensive analysis of risks and timely response, and reduces the efficiency and effectiveness of pathological critical value management.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention proposes a pathology critical value warning method and system based on a pathology knowledge graph, which solves the problems mentioned in the above background technology that the existing technology is not convenient for identifying key feature data, cannot effectively screen out the data that has the greatest impact on risk assessment, is not convenient for building a warning model, cannot evaluate the risk level of test indicators through the warning model, is not convenient for early detection of potential risks, and is not convenient for converting the risk trend of test indicators at future times into actual risk assessment, thereby reducing the warning efficiency. At the same time, it is not convenient to analyze the risk level of test indicators, judge the key factors affecting the risk level, and issue warning reminders, which makes it impossible to achieve a comprehensive analysis and timely response to risks, thereby reducing the efficiency and effectiveness of pathology critical value management.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0008] According to one aspect of the present invention, a pathology critical value early warning method based on a pathology knowledge graph is provided, and the pathology critical value early warning method based on a pathology knowledge graph comprises the following steps:

[0009] S1. Obtain the test index data related to the pathological critical value in the pathological knowledge graph, pre-process the test index data, and obtain the test index characteristic data;

[0010] S2. Using the feature classification algorithm to analyze the inspection index feature data, identify key feature data;

[0011] S3. Based on key feature data, build an early warning model and evaluate the risk level of inspection indicators through the early warning model;

[0012] S4. Analyze the risk level of the inspection indicators through the level analysis algorithm, determine the key factors affecting the risk level of the inspection indicators, and issue early warning reminders based on the key factors;

[0013] Among them, based on the key feature data, building an early warning model and evaluating the risk level of the inspection indicators through the early warning model include the following steps:

[0014] S31. Convert the obtained key feature data into a risk assessment cycle;

[0015] S32, replacing the risk assessment cycle into the cycle index adjustment algorithm, dividing the obtained key feature data into cycle indicators and trend indicators, and constructing and training an early warning model through the cycle indicators and trend indicators to predict the future inspection indicator risk trend;

[0016] S33. Use the mean absolute percentage error as the evaluation criterion for the accuracy of the early warning model;

[0017] S34. Based on the model optimization algorithm, adjust and optimize the parameters of the early warning model according to the trend indicators, and find the best parameter configuration;

[0018] S35. Apply the optimal parameter configuration to the constructed early warning model, use the second-order adaptive coefficient method to predict the trend data, and obtain the risk trend of the inspection indicator at the future moment;

[0019] S36. Convert the obtained inspection indicator risk trend at future moments into actual inspection indicator risk assessment, construct the final early warning model, and evaluate the inspection indicator risk level.

[0020] Further, obtaining the test index data related to the pathological critical value in the pathological knowledge graph, preprocessing the test index data, and obtaining the test index feature data includes the following steps:

[0021] S11, obtaining duplicate data, missing values ​​and abnormal values ​​of the test index data related to the pathological critical value, and performing denoising, filtering and smoothing on the duplicate data, missing values ​​and abnormal values;

[0022] S12, connecting the unprocessed data in the inspection index data to generate a new data table, associating different data tables through external key values, generating a complete data table, and obtaining an accurate data set;

[0023] S13. Extract features from the accurate data set and obtain test indicator feature data.

[0024] Further, using the inspection index characteristic data obtained by the characteristic classification algorithm to analyze, identifying the key characteristic data includes the following steps:

[0025] S21, setting a test index characteristic data sample set, and presetting the number of mutation and crossover operations;

[0026] S22, mutating the test indicator feature data sample set through a mutation operation, calculating the difference vector between each test indicator feature data sample, determining the best feature data sample in the test indicator feature data sample set, and using the best feature data sample as a new mutated test indicator feature data sample;

[0027] S23, using a feature classification algorithm to perform a new cluster analysis on the test indicator feature data samples, to determine the cluster to which each test indicator feature data sample belongs and the optimal distance between each test indicator feature data sample;

[0028] S24, analyzing the formed new test index characteristic data sample set, comparing the test index characteristic data sample after mutation crossover with the objective function, and selecting the test index characteristic data sample set with the best objective function value as the new test index characteristic data sample set;

[0029] S25. Determine whether the preset termination condition is met. If so, the evolution is terminated, and the current best test indicator feature data sample set is determined as the key feature data of the test indicator.

[0030] Further, using the feature classification algorithm to perform a new cluster analysis on the test indicator feature data samples, determining the cluster where each test indicator feature data sample is located and the optimal distance between each test indicator feature data sample includes the following steps:

[0031] S231, randomly selecting a number of test indicator characteristic data samples from the test indicator characteristic data sample set as initial cluster center points;

[0032] S232, assigning each unassigned test index characteristic data sample to the cluster center point closest to it;

[0033] S233, randomly selecting a non-cluster center test indicator feature data sample from the test indicator feature data sample set, and recording it as a random test indicator feature data sample;

[0034] S234, calculating the total cost change caused by replacing the current cluster center point with the random inspection indicator feature data sample, and updating the cluster center set;

[0035] S235. When the cluster center point no longer changes, determine the final clustering of each test indicator feature data sample, and calculate the optimal distance between each test indicator feature data sample.

[0036] Furthermore, the formula for mutating the test index feature data sample set through mutation operation is:

[0037] ;

[0038] In the formula, Expressed as Moment, c A sample of characteristic data of the variation test indicators;

[0039] Represented as parent basis vectors;

[0040] is the parent difference vector;

[0041] S Expressed as a scaling factor;

[0042] t 1. t 2. t3 represent three different indexes randomly selected from the index of the current test index feature data sample set.

[0043] Furthermore, based on the model optimization algorithm, adjusting and optimizing the parameters of the early warning model according to the trend indicators and finding the best parameter configuration include the following steps:

[0044] S341, initializing key parameters of the model optimization algorithm;

[0045] S342, randomly initialize the parameter configuration of the early warning model, calculate the initial risk assessment value of each parameter configuration, and use it as the maximum risk indicator of the parameter;

[0046] S343. Calculate the relative importance of each risk indicator, and determine the direction of adjusting the early warning model parameters according to the relative importance;

[0047] S344. Update the early warning model parameter configuration according to the relative importance and trend indicators, and randomly perturb the optimal parameter configuration;

[0048] S345. Re-evaluate the accuracy and effectiveness of the model based on the updated parameter configuration;

[0049] S346. If the maximum number of iterations is reached, stop the optimization; otherwise, continue iterating and adjust the parameter configuration;

[0050] S347. When the optimization process is completed, the best parameter configuration and the optimized early warning model are output.

[0051] Furthermore, the formula for updating the early warning model parameter configuration is:

[0052] ;

[0053] in, Indicates the current parameter configuration;

[0054] Represented as the optimal parameter configuration;

[0055] Expressed as step size factor;

[0056] It is expressed as the attractiveness coefficient;

[0057] Represented as a random number.

[0058] Further, the risk level of the inspection indicator is analyzed by the level analysis algorithm, the key factors affecting the risk level of the inspection indicator are determined, and early warning reminders are issued according to the key factors, including the following steps:

[0059] S41. Obtain risk levels of different inspection indicators and compare them with industry standard risk assessment indicators;

[0060] S42. Analyze the key factors that affect the risk level of inspection indicators and set the corresponding risk assessment level;

[0061] S43, using a grade analysis algorithm to train the obtained risk levels of different inspection indicators, calculate the risk score, and determine whether it belongs to a high risk level;

[0062] S44, normalizing the risk score obtained by the grade analysis algorithm and calculating the fuzzy set;

[0063] S45. Analytical experts evaluate each key factor at different risk levels to form a risk assessment matrix;

[0064] S46. Using fuzzy reasoning to calculate the normalized risk score and the risk assessment matrix to obtain a comprehensive risk assessment result;

[0065] S47. Determine the relative position of the inspection indicators in different risk levels based on the component values ​​of the comprehensive evaluation results and determine the risk level;

[0066] S48, determining whether the risk level exceeds a set threshold, if so, marking it as high risk, if not, treating it as low or medium risk;

[0067] S49. Based on the results of the grade analysis algorithm and the advice of analysis experts, determine the key factors that affect the risk level of the inspection indicators and issue warnings based on the key factors.

[0068] Further, analyzing the key factors that affect the risk level of the inspection indicators and setting the corresponding risk assessment level includes the following steps:

[0069] S421. Collect various key factors that affect the risk level of inspection indicators based on industry research and inspection data;

[0070] S422. Screen and classify the key factors that affect the risk level of various inspection indicators collected, remove duplicate factors, and summarize them into a set of measurable key factors;

[0071] S423. Set weights for each key factor affecting the risk level based on industry research and inspection experience, and set risk assessment levels for each key factor;

[0072] S424. Reflect the importance and performance of each key factor in the risk assessment of the inspection indicator based on the set risk assessment level.

[0073] According to another aspect of the present invention, a pathological critical value warning system based on a pathological knowledge graph is also provided. The pathological critical value warning system based on a pathological knowledge graph comprises:

[0074] A data acquisition module is used to obtain the test index data related to the pathological critical value in the pathological knowledge graph, pre-process the test index data, and obtain the test index characteristic data;

[0075] A feature recognition module is used to identify key feature data by analyzing the inspection index feature data obtained by the feature classification algorithm;

[0076] Model building module, used to build early warning models based on key feature data, and evaluate the risk level of inspection indicators through early warning models;

[0077] The data analysis module is used to analyze the risk level of the inspection indicators through the level analysis algorithm, determine the key factors affecting the risk level of the inspection indicators, and issue early warning reminders based on the key factors;

[0078] Among them, the data acquisition module is connected with the model construction module through the feature recognition module, and the model construction module is connected with the data analysis module.

[0079] The beneficial effects of the present invention are:

[0080] 1. The present invention can ensure the quality and consistency of data by obtaining the test indicator data in the pathology knowledge graph and performing preprocessing. It can effectively screen out the data that has the greatest influence on risk assessment by using the feature classification algorithm to identify key feature data, and build an early warning model based on the key feature data. The risk level of the test indicator is evaluated by the early warning model, which helps to discover potential risks early. The risk level of the test indicator is analyzed by the level analysis algorithm, the key factors affecting the risk level are determined, and early warning reminders are issued, which can achieve comprehensive analysis and timely response to risks, thereby improving the efficiency and effectiveness of pathology critical value management.

[0081] 2. The present invention ensures the diversity and representativeness of the data sample set by presetting the number of mutation and crossover operations, improves the quality of the feature data sample set by using mutation operations and differential vector calculations, ensures the accurate extraction of key feature data, and determines the clustering affiliation and optimal distance of each test indicator feature data sample through cluster analysis, thereby achieving effective classification of data samples and facilitating further risk assessment and prediction.

[0082] 3. The present invention constructs an early warning model by utilizing key feature data, and conducts training and optimization to evaluate the risk level of test indicators. By constructing periodic and trend indicators, it accurately predicts future risk trends and provides forward-looking early warnings. It evaluates model accuracy using mean absolute percentage error, optimizes model parameters, ensures the accuracy of prediction results, and achieves high-precision prediction of future test indicator risk trends, thereby achieving efficient construction and optimization of the early warning model and ensuring the reliability of risk assessment. Through sophisticated feature data extraction, efficient feature classification and clustering analysis, dynamic optimization iteration and accurate risk assessment, it achieves effective early warning of pathological critical values.

[0083] 4. The present invention ensures the scientificity and comprehensiveness of the evaluation process through detailed data comparison and key factor analysis, combines expert evaluation with fuzzy reasoning to provide authoritative and reliable risk assessment results, ensures timely discovery and response to potential risks through detailed threshold judgment and early warning reminders, and optimizes the pathological critical value management process through comprehensive risk level judgment and key factor analysis, thereby improving medical quality and patient safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0085] Figure 1 is a flow chart of a pathology critical value warning method based on a pathology knowledge graph according to an embodiment of the present invention;

[0086] Figure 2 It is a principle block diagram of a pathology critical value warning system based on a pathology knowledge graph according to an embodiment of the present invention.

[0087] In the figure:

[0088] 1. Data acquisition module; 2. Feature recognition module; 3. Model building module; 4. Data analysis module. DETAILED DESCRIPTION

[0089] The technical solutions in the embodiments of the present invention will be described clearly and completely 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.

[0090] In the description of the present invention, unless otherwise specified, "plurality" means two or more. In addition, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0091] According to an embodiment of the present invention, a pathology critical value warning method and system based on a pathology knowledge graph are provided.

[0092] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to the pathological critical value warning method based on the pathological knowledge graph according to an embodiment of the present invention, the pathological critical value warning method based on the pathological knowledge graph includes the following steps:

[0093] S1. Obtain the test index data related to the pathological critical value in the pathological knowledge graph, pre-process the test index data, and obtain the test index characteristic data;

[0094] It needs to be explained that the pathology knowledge graph is a systematic and structured form of knowledge representation. It combines a large amount of knowledge in the field of pathology and expresses various pathology-related entities (such as diseases, symptoms, test indicators, treatment methods, etc.) and their relationships through a graph structure (nodes and edges).

[0095] Specifically, the pathology knowledge graph includes:

[0096] 1) Entities (Nodes):

[0097] Disease: includes various pathological conditions, such as heart disease, liver disease, diabetes, etc.

[0098] Symptoms: such as fever, pain, fatigue, etc.

[0099] Test indicators: such as hemoglobin, white blood cell count, troponin, etc.

[0100] Treatment methods: such as drug therapy, surgical treatment, physical therapy, etc.

[0101] 2) Edges:

[0102] Disease-symptom relationships: Describe the symptoms that a particular disease may cause.

[0103] Relationship between disease and test index: Describe the abnormal test index that may be caused by a specific disease.

[0104] Relationship between symptoms and test indicators: Describe how specific symptoms may be reflected in abnormal test indicators.

[0105] Disease-treatment relationships: Describe conventional treatments for a particular disease and their effectiveness.

[0106] It should be explained that the test index data related to pathological critical values ​​refer to those pathological test indicators that are considered to be highly important and urgent in clinical practice. These indicators are usually used to monitor and evaluate the health status of patients. Once abnormal values ​​appear, it may mean that the patient is in an emergency or critical state and requires immediate medical intervention.

[0107] Specifically, the inspection index data include:

[0108] 1) Hematological indicators:

[0109] Hemoglobin (Hb): Levels below or above the normal range may indicate anemia or polycythemia.

[0110] Red Blood Cell Count (RBC): An abnormal number may indicate anemia, bleeding, bone marrow problems, etc.

[0111] White Blood Cell Count (WBC): A high WBC count may indicate infection, inflammation, or leukemia, while a low WBC count may indicate immune deficiency or bone marrow suppression.

[0112] Platelet count (PLT): Low platelet count may lead to bleeding risk, while high platelet count may indicate a myeloproliferative disorder.

[0113] 2) Biochemical indicators:

[0114] Glucose: Abnormal blood sugar levels may indicate diabetes or hypoglycemia.

[0115] Potassium (K): Hyperkalemia or hypokalemia may cause serious arrhythmias.

[0116] Calcium (Ca): Abnormal blood calcium levels may affect nerve and muscle function.

[0117] Creatinine: High creatinine levels may indicate renal insufficiency.

[0118] 3) Enzyme indicators:

[0119] Alanine Aminotransferase (ALT): High levels may indicate liver cell damage.

[0120] Aspartate Aminotransferase (AST): High levels may indicate liver, heart, or muscle damage.

[0121] 4) Blood gas analysis:

[0122] Arterial oxygen partial pressure (PaO2): Low oxygen partial pressure may indicate respiratory failure.

[0123] Partial Pressure of Carbon Dioxide (PaCO2): High PaCO2 may indicate respiratory acidosis.

[0124] 5) Coagulation function:

[0125] International Normalized Ratio (INR): An abnormal INR may indicate a coagulation disorder.

[0126] Activated Partial Thromboplastin Time (APTT): Prolongation may indicate coagulation disorder or excessive anticoagulant therapy.

[0127] 6) Cardiac markers:

[0128] Troponin: High levels may indicate a myocardial infarction.

[0129] B-type Natriuretic Peptide (BNP): High levels may indicate heart failure.

[0130] S2. Using the feature classification algorithm to analyze the inspection index feature data, identify key feature data;

[0131] S3. Based on key feature data, build an early warning model and evaluate the risk level of inspection indicators through the early warning model;

[0132] It should be explained that the risk level of test indicators refers to the evaluation of the measurement results of a specific test indicator and the classification of the risk it may cause to the patient's health, including:

[0133] Normal Range:

[0134] The indicator results are within the normal range, indicating that the patient is in good health and does not require special treatment.

[0135] Mild Abnormality:

[0136] Results that are slightly above or below the normal range may require further monitoring or minor intervention but are generally not immediately life-threatening.

[0137] Moderate Abnormality:

[0138] If the indicator results are significantly higher or lower than the normal range, it may indicate a health problem that requires further examination and treatment.

[0139] Severe Abnormality:

[0140] Results that deviate significantly from the normal range may reflect a serious health problem that requires immediate medical attention.

[0141] Critical Value:

[0142] Extremely abnormal results indicate that the patient may be in a life-threatening condition and requires emergency medical intervention.

[0143] S4. Analyze the risk level of the inspection indicators through the level analysis algorithm, determine the key factors affecting the risk level of the inspection indicators, and issue early warning reminders based on the key factors;

[0144] It should be noted that the key factors include:

[0145] 1) Basic information of the patient:

[0146] Age: The normal range of test indicators may be different for patients of different age groups, and the elderly and children need special attention.

[0147] Gender: Some test indicators differ between genders.

[0148] 2) Medical history and health status:

[0149] Past medical history: including chronic diseases, acute diseases, family genetic diseases, etc., have a significant impact on certain test indicators.

[0150] Current medical condition: Whether you currently have a disease or are receiving treatment, which will affect the interpretation of certain indicators.

[0151] 3) Lifestyle and behavior:

[0152] Eating habits: A high-salt, high-sugar, and high-fat diet may affect blood sugar, blood lipids and other indicators.

[0153] Exercise habits: Lack of exercise or excessive exercise will affect certain body indicators.

[0154] Smoking and drinking: These behaviors have a significant impact on blood, liver function and other indicators.

[0155] 4) Drug use:

[0156] Drug treatment: the type, dosage and duration of medication currently being used, especially those that affect blood, liver and kidney function.

[0157] 5) Characteristics of the test indicator itself:

[0158] Volatility of test indicators: Some indicators such as blood sugar fluctuate greatly during the day, and the impact of measurement time needs to be considered.

[0159] Test methods and accuracy: Test methods and equipment may vary between laboratories, and these technical factors need to be considered.

[0160] Among them, based on the key feature data, building an early warning model and evaluating the risk level of the inspection indicators through the early warning model include the following steps:

[0161] S31. Convert the obtained key feature data into a risk assessment cycle;

[0162] S32, replacing the risk assessment cycle into the cycle index adjustment algorithm, dividing the obtained key feature data into cycle indicators and trend indicators, and constructing and training an early warning model through the cycle indicators and trend indicators to predict the future inspection indicator risk trend;

[0163] S33. Use the mean absolute percentage error as the evaluation criterion for the accuracy of the early warning model;

[0164] S34. Based on the model optimization algorithm, adjust and optimize the parameters of the early warning model according to the trend indicators, and find the best parameter configuration;

[0165] S35. Apply the optimal parameter configuration to the constructed early warning model, use the second-order adaptive coefficient method to predict the trend data, and obtain the risk trend of the inspection indicator at the future moment;

[0166] S36. Convert the obtained inspection indicator risk trend at future moments into actual inspection indicator risk assessment, construct the final early warning model, and evaluate the inspection indicator risk level.

[0167] It should be explained that, taking the blood glucose level monitoring of diabetic patients in a certain hospital as an example, the blood glucose level data of diabetic patients are first collected, including the daily morning fasting blood glucose value and the daily blood glucose value 2 hours after dinner. The monthly data is used as an evaluation cycle, and the key feature data of each cycle is extracted, such as the monthly average blood glucose value, the monthly maximum blood glucose value, the monthly blood glucose fluctuation range, etc. The monthly average blood glucose value, the maximum blood glucose value, etc. are defined as periodic indicators, and the blood glucose change trend in multiple cycles is calculated, such as the blood glucose change rate within 3 months. Using periodic indicators and trend indicators, appropriate algorithms (such as LSTM neural networks, ARIMA models, random forests) are selected to construct and train early warning models to predict future trends in blood glucose level changes. The predicted values ​​of the early warning model are compared with the actual values, and the prediction error is calculated by the mean absolute percentage error (MAPE). MAPE is a commonly used indicator to measure the deviation between the predicted value and the actual value. MAPE provides a An intuitive error percentage makes the performance evaluation of the early warning model clearer, easier to understand and compare, evaluates the prediction accuracy of the model according to the size of MAPE, adjusts the early warning model parameters to improve accuracy, uses model optimization algorithms such as firefly algorithm, genetic algorithm or gradient descent method, adjusts the early warning model parameters according to trend indicators, finds the best parameter configuration to improve the prediction accuracy and responsiveness of the early warning model, and continuously optimizes the model parameters according to the trend data of blood glucose levels to make the prediction results more accurate. The optimized best parameter configuration is applied to the early warning model, and the trend data is predicted using the second-order adaptive coefficient method. For example, the Holt-Winters method is used for time series prediction, and the predicted future blood glucose level trend is converted into a specific risk level (such as normal, mildly abnormal, moderately abnormal, severely abnormal, critical). According to the converted risk level, the patient's blood glucose level is risk assessed and an early warning reminder is generated.

[0168] Preferably, obtaining the test index data related to the pathological critical value in the pathological knowledge graph, preprocessing the test index data, and obtaining the test index feature data includes the following steps:

[0169] S11, obtaining duplicate data, missing values ​​and abnormal values ​​of the test index data related to the pathological critical value, and performing denoising, filtering and smoothing on the duplicate data, missing values ​​and abnormal values;

[0170] S12, connecting the unprocessed data in the inspection index data to generate a new data table, associating different data tables through external key values, generating a complete data table, and obtaining an accurate data set;

[0171] S13. Extract features from the accurate data set and obtain test indicator feature data.

[0172] It needs to be explained that, taking the blood glucose level monitoring of diabetic patients in a certain hospital as an example, the blood glucose data of diabetic patients are classified according to date and time period, the blood glucose values ​​in the morning and after dinner every day are calculated, the duplicate, missing and abnormal data are detected and processed to ensure data quality, the blood glucose data in the morning and after dinner are connected to generate a complete data table to ensure the accuracy and consistency of the data, and key features such as average, maximum value, fluctuation range, etc. are extracted from the complete data set, which are standardized to generate a feature data set.

[0173] Preferably, identifying key feature data using the inspection index feature data obtained by feature classification algorithm analysis includes the following steps:

[0174] S21, setting a test index characteristic data sample set, and presetting the number of mutation and crossover operations;

[0175] S22, mutating the test indicator feature data sample set through a mutation operation, calculating the difference vector between each test indicator feature data sample, determining the best feature data sample in the test indicator feature data sample set, and using the best feature data sample as a new mutated test indicator feature data sample;

[0176] Specifically, the mutation operation is the processing of the mutation operation in the differential evolution algorithm (DE).

[0177] S23, using a feature classification algorithm to perform a new cluster analysis on the test indicator feature data samples, to determine the cluster to which each test indicator feature data sample belongs and the optimal distance between each test indicator feature data sample;

[0178] S24, analyzing the formed new test index characteristic data sample set, comparing the test index characteristic data sample after mutation crossover with the objective function, and selecting the test index characteristic data sample set with the best objective function value as the new test index characteristic data sample set;

[0179] Specifically, the crossover operation is the processing of the crossover operation in the differential evolution algorithm (DE).

[0180] S25. Determine whether the preset termination condition is met. If so, the evolution is terminated, and the current best test indicator feature data sample set is determined as the key feature data of the test indicator.

[0181] Specifically, the feature classification algorithm is a K-medoids clustering algorithm based on differential evolution. The idea of ​​the K-medoids clustering algorithm based on differential evolution is: first, a group of a certain size is randomly selected from the population, and the size of the group is set to a fixed value. At the same time, the maximum size limit of the population is set. Then, the differential evolution algorithm (DE) is used to perform mutation and crossover operations on this population to determine the optimal mutant individuals and experimental individuals in the population, thereby avoiding the population from falling into the local optimum and enhancing the algorithm's local search ability in the cluster space area. Next, according to the K-medoids clustering algorithm, a new clustering analysis is performed on these optimized experimental individuals to determine the cluster to which each individual belongs and the optimal distance between them. In this way, the individual with the lowest objective function value can be found and used as a member of the new population, and the optimal solution is finally output. Finally, the global search ability of the differential evolution algorithm (DE) and the local search efficiency of the K-medoids clustering algorithm in a specific spatial area are combined to achieve more efficient clustering analysis and feature extraction.

[0182] Specifically, the differential evolution algorithm (DE) is an algorithm based on population evolution. It has the characteristics of memorizing individual optimal solutions and sharing information within the population. That is, it solves the optimization problem through cooperation and competition among individuals in the population. Its essence is a greedy genetic algorithm based on real number coding with the idea of ​​preserving excellence.

[0183] It should be explained that, taking the blood glucose level monitoring of diabetic patients in a hospital as an example, the blood glucose level data of diabetic patients are collected, including the fasting blood glucose value every morning and the blood glucose value 2 hours after dinner. The extracted feature data include: mean, maximum, minimum, standard deviation, rate of change, etc. The number of mutation and crossover operations is set, for example, each operation is performed 50 times, and each mutated sample is cross-operated with its original sample. According to the crossover rate, it is determined whether to take values ​​from the mutated sample, generate a new candidate sample set, compare the crossover sample with the original sample, and select the better sample as the new generation sample set according to the fitness (such as blood glucose control effect, daily quality of life of patients). The fitness evaluation criteria can include the stability and fluctuation range of blood glucose values, and the updated sample set is clustered using the K-medoids clustering algorithm based on differential evolution to find patients with similar blood glucose control patterns. The cluster center is re-determined based on the optimal distance, and the samples are assigned to the nearest cluster. The mutation, crossover and selection steps are repeated until the preset termination conditions are reached, such as reaching the maximum number of iterations or the fitness improvement is not significant. The optimal blood glucose monitoring feature data sample set is finally used to adjust the treatment plan for diabetic patients, such as adjusting the timing and dosage of insulin injections, optimizing diet and exercise plans, so as to improve blood glucose control and the quality of life of patients.

[0184] Preferably, using a feature classification algorithm to perform a new cluster analysis on the test indicator feature data samples, and determining the cluster where each test indicator feature data sample belongs and the optimal distance between each test indicator feature data sample includes the following steps:

[0185] S231, randomly selecting a number of test indicator characteristic data samples from the test indicator characteristic data sample set as initial cluster center points;

[0186] S232, assigning each unassigned test index characteristic data sample to the cluster center point closest to it;

[0187] S233, randomly selecting a non-cluster center test indicator feature data sample from the test indicator feature data sample set, and recording it as a random test indicator feature data sample;

[0188] S234, calculating the total cost change caused by replacing the current cluster center point with the random inspection indicator feature data sample, and updating the cluster center set;

[0189] S235. When the cluster center point no longer changes, determine the final clustering of each test indicator feature data sample, and calculate the optimal distance between each test indicator feature data sample.

[0190] It should be explained that, taking the blood glucose level monitoring of diabetic patients in a hospital as an example, several samples are randomly selected from the blood glucose monitoring feature data sample set of diabetic patients as the initial cluster centers. These features may include blood glucose fluctuation amplitude, postprandial blood glucose peak, daytime blood glucose changes, patient diet and exercise habits, etc. The selection of the initial cluster center can be based on simple random sampling, or a more complex heuristic method can be used to ensure the diversity of cluster centers. Each unassigned blood glucose monitoring feature data sample is traversed and assigned to the cluster center point with the closest Euclidean distance to it. This step ensures that each sample belongs to the cluster that is most likely to represent its blood glucose control mode, thereby improving the accuracy and usefulness of the clustering results. A sample that is not a cluster center is randomly selected from the sample set as a random sample to examine the total cost change caused by replacing it with the current cluster center. The total cost calculation can be based on the compactness of the samples within the cluster (i.e., the average distance between the samples within the cluster and the center point). If the total cost after replacement is reduced (i.e., the clustering is tighter), the cluster center set is updated. When the cluster center points are stable and no longer change after several iterations, the final clustering of each blood glucose monitoring feature data sample is confirmed, and the optimal distance between each feature data sample is calculated. This can usually be achieved by calculating the average distance or minimum distance between samples within the cluster.

[0191] Preferably, the formula for mutating the test index feature data sample set through mutation operation is:

[0192] ;

[0193] In the formula, Expressed as Moment, c A sample of characteristic data of the variation test indicators;

[0194] Represented as parent basis vectors;

[0195] Specifically, in m At the moment, it is used as a parent test indicator feature data sample of the basis vector. This is the benchmark point used to generate mutant individuals.

[0196] is the parent difference vector;

[0197] Specifically, in m At time t, two different parent test indicator feature data samples are used to create a difference vector, and the parent difference vector represents the difference between the two samples.

[0198] S Expressed as a scaling factor;

[0199] Specifically, the scaling factor is used to control the intensity of the influence of the differential vector on the mutant individual. Its value is usually between 0.4 and 0.6. When the scaling factor is close to 0.5, the mutant individual will be closer to the parent basis vector.

[0200] t 1. t 2. t 3 represent three different indexes randomly selected from the index of the current test index feature data sample set.

[0201] Preferably, based on the model optimization algorithm, adjusting and optimizing the parameters of the early warning model according to the trend indicators, and finding the optimal parameter configuration includes the following steps:

[0202] S341, initializing key parameters of the model optimization algorithm;

[0203] It needs to be explained that the application example is: choose an algorithm suitable for parameter optimization, such as genetic algorithm, firefly algorithm or particle swarm optimization (PSO), and initialize its key parameters such as population size, number of iterations, crossover rate and mutation rate (genetic algorithm) or particle speed and position (PSO).

[0204] Theoretical basis: Initialization parameter setting is the basis of the optimization process. Correct parameter setting can effectively improve the optimization efficiency and success rate.

[0205] S342, randomly initialize the parameter configuration of the early warning model, calculate the initial risk assessment value of each parameter configuration, and use it as the maximum risk indicator of the parameter;

[0206] It should be explained that the application example is: a set of initial parameter configurations are randomly generated in the parameter space, each parameter configuration represents a possible solution. The risk assessment value of the early warning model under each parameter configuration is calculated and used as the basis for evaluating the quality of the parameter configuration.

[0207] Theoretical basis: Random initialization helps explore the parameter space, ensures the diversity of solutions, and prevents the algorithm from converging to the local optimal solution too early.

[0208] S343. Calculate the relative importance of each risk indicator, and determine the direction of adjusting the early warning model parameters according to the relative importance;

[0209] It should be explained that the application example is: using methods such as sensitivity analysis or impact assessment to calculate the influence of each parameter on the output of the early warning model (risk assessment value) to determine which parameters are more critical.

[0210] Theoretical basis: Relative importance assessment can help determine which parameters should be adjusted first during the optimization process to more effectively improve the early warning model performance.

[0211] S344. Update the early warning model parameter configuration according to the relative importance and trend indicators, and randomly perturb the optimal parameter configuration;

[0212] It is necessary to explain that the application example is: according to the relative importance of each parameter and the observed trend, the parameter value is adjusted. This step can be combined with random perturbations to prevent the model optimization algorithm from falling into the local optimum.

[0213] Theoretical basis: Parameter updating and perturbation are key strategies to maintain optimization dynamics and avoid local optimality.

[0214] S345. Re-evaluate the accuracy and effectiveness of the model based on the updated parameter configuration;

[0215] It should be explained that the application example is: using the updated parameter configuration, re-running the early warning model, and using pre-set evaluation indicators (such as MAPE) to evaluate the accuracy and effectiveness of the early warning model.

[0216] S346. If the maximum number of iterations is reached, stop the optimization; otherwise, continue iterating and adjust the parameter configuration;

[0217] It should be explained that the application example is: repeatedly executing the parameter update and early warning model evaluation process until the set maximum number of iterations is reached or the performance of the early warning model is no longer significantly improved.

[0218] Theoretical basis: The iterative process helps to fine-tune the parameters and gradually approach the optimal solution.

[0219] S347. When the optimization process is completed, the best parameter configuration and the optimized early warning model are output.

[0220] It should be explained that in the application example, after the optimization process is completed, the parameter configuration that achieves the best performance index and the corresponding early warning model version are output, and this version will be used for the actual risk index evaluation.

[0221] Theoretical basis: The final output warning model should have high accuracy and reliability, and be able to support actual inspection indicator risk level safety management decisions.

[0222] Specifically, the model optimization algorithm is the firefly algorithm, which contains two elements, namely brightness and attractiveness. Brightness reflects the pros and cons of the firefly's position and determines its moving direction, and attractiveness determines the distance the firefly moves. By continuously updating brightness and attractiveness, the target optimization is achieved.

[0223] Preferably, the formula for updating the early warning model parameter configuration is:

[0224] ;

[0225] in, Indicates the current parameter configuration;

[0226] Represented as the optimal parameter configuration;

[0227] Expressed as step size factor;

[0228] Specifically, the step factor is a [0, 1] constant that is used to determine the scale of random perturbations and help the model optimization algorithm escape from the local optimal solution.

[0229] It is expressed as the attractiveness coefficient;

[0230] Specifically, the attractiveness coefficient determines the degree to which the current parameter configuration moves toward a better parameter configuration during parameter update.

[0231] Represented as a random number.

[0232] Specifically, is a random number uniformly distributed on [0, 1], which is used to introduce randomness and increase the diversity of the search.

[0233] Specifically, in the formula for updating the early warning model parameter configuration, use The purpose is to introduce randomness, which can help the algorithm avoid falling into the local optimal solution and enable more extensive exploration in the search space.

[0234] It needs to be explained that rand is a uniformly distributed random number in the range [0, 1], and its expected value is 0.5. rand When 0.5 is subtracted from The range of becomes [-1 / 2, 1 / 2], and its expected value becomes 0. After this processing, the center position of the random number is near 0, which allows the algorithm to have the opportunity to move towards the "good" solution during the search process, as well as the opportunity to move away from the current position and explore other possible solutions. In the firefly algorithm, the step size factor and Multiplication means that the algorithm will introduce a certain degree of random exploration in each update step, which helps the algorithm to jump out of the local optimal solution and find the global optimal solution. The expected value of is 0. In the long run, this randomness will not have a negative impact on the convergence trend of the algorithm.

[0235] Preferably, analyzing the risk level of the inspection indicator by a level analysis algorithm, determining the key factors affecting the risk level of the inspection indicator, and issuing early warning reminders based on the key factors include the following steps:

[0236] S41. Obtain risk levels of different inspection indicators and compare them with industry standard risk assessment indicators;

[0237] S42. Analyze the key factors that affect the risk level of inspection indicators and set the corresponding risk assessment level;

[0238] S43, using a grade analysis algorithm to train the obtained risk levels of different inspection indicators, calculate the risk score, and determine whether it belongs to a high risk level;

[0239] Specifically, using the level analysis algorithm to train the obtained different inspection indicator risk levels, calculate the risk score, and determine whether it belongs to the high risk level includes the following steps:

[0240] S431, dividing the identified key factors affecting the risk level of the inspection indicators into a training set and a test set;

[0241] S432, using the training set data, training a hierarchical analysis model (i.e., an isolation forest model) by randomly selecting split points within the range of features and feature values;

[0242] S433, construct multiple classification trees to form a complete hierarchical analysis model;

[0243] S434, using the grade analysis model, calculating the risk score of each key factor data point in the test set that affects the risk level of the inspection indicator;

[0244] S435. Based on the risk score, a risk level is calculated for each key factor data affecting the risk level of the inspection indicator, and the risk levels of different inspection indicators are measured;

[0245] S436. Based on the set threshold, use the calculated risk score to determine whether each key factor data affecting the risk level of the inspection indicator belongs to a high risk level.

[0246] It should be explained that during the training process, the data points are divided into two subsets according to the selected eigenvalue (one containing data points less than or equal to the eigenvalue, and the other containing data points greater than the eigenvalue), and a split operation is performed; this process is recursively repeated for each subset until the stopping condition is met (such as the subset size reaches a predetermined threshold and the depth of the tree reaches the maximum value).

[0247] S44, normalizing the risk score obtained by the grade analysis algorithm and calculating the fuzzy set;

[0248] S45. Analytical experts evaluate each key factor at different risk levels to form a risk assessment matrix;

[0249] S46. Using fuzzy reasoning (i.e., fuzzy operation operator) to calculate the normalized risk score (i.e., weight vector) and the risk assessment matrix (i.e., fuzzy relationship matrix) to obtain a comprehensive risk assessment result;

[0250] S47. Determine the relative position of the inspection indicators in different risk levels based on the component values ​​of the comprehensive evaluation results and determine the risk level;

[0251] S48, determining whether the risk level exceeds a set threshold, if so, marking it as high risk, if not, treating it as low or medium risk;

[0252] S49. Based on the results of the grade analysis algorithm and the advice of analysis experts, determine the key factors that affect the risk level of the inspection indicators and issue warnings based on the key factors.

[0253] It should be explained that, taking the blood sugar level monitoring of diabetic patients in a hospital as an example, first, key factors (such as hyperglycemia, hypoglycemia, abnormal blood sugar fluctuations, etc.) are identified from the patient's real-time blood sugar monitoring and historical records, and these factors are compared with various factors known to affect blood sugar levels. This can be done by analyzing the common blood sugar abnormalities recorded in the database. Key factors are graded according to their severity and frequency, for example, blood sugar abnormalities are divided into mild, moderate, and severe. These levels reflect the likelihood of blood sugar abnormalities and potential health effects. The identified key factor data are trained using a hierarchical analysis algorithm to calculate the risk scores for various types of blood sugar abnormalities that may occur in diabetic patients. For example, a logistic regression model is used to predict the probability of dysglycemia based on the severity of factors, and the obtained risk scores are normalized so that all scores fall on a uniform scale. Fuzzy sets are calculated, such as using triangular fuzzy numbers to represent risk levels. Professionals evaluate different risk levels based on various factors that affect dysglycemia to form a detailed risk assessment matrix. Fuzzy reasoning is used to integrate the normalized risk scores with the risk assessment matrix to obtain a comprehensive risk assessment result for each dysglycemia. The relative position of dysglycemia in different risk levels is judged based on the component values ​​of the comprehensive risk assessment results, and the severity of each type of abnormality is determined, so that high-risk dysglycemia can be given priority.

[0254] It should be explained that the rank analysis algorithm is a fuzzy isolation forest algorithm, which is a fuzzy improved algorithm based on the isolation forest. Isolation forest is a very effective anomaly detection method. Its core idea is to use a binary tree structure to isolate data points, in which the positions of normal data and abnormal data in the tree are significantly different, and abnormal data are usually isolated earlier. In the present invention, the purpose is to normalize the risk score obtained and construct a fuzzy set to handle the uncertainty and ambiguity in the key factor data that affects the risk level of the inspection indicator.

[0255] Preferably, analyzing the key factors affecting the risk level of the inspection indicators and setting the corresponding risk assessment level includes the following steps:

[0256] S421. Collect various key factors that affect the risk level of inspection indicators based on industry research and inspection data;

[0257] S422. Screen and classify the key factors that affect the risk level of various inspection indicators collected, remove duplicate factors, and summarize them into a set of measurable key factors;

[0258] S423. Set weights for each key factor affecting the risk level based on industry research and inspection experience, and set risk assessment levels for each key factor;

[0259] S424. Reflect the importance and performance of each key factor in the risk assessment of the inspection indicator based on the set risk assessment level.

[0260] It needs to be explained that, taking the blood sugar level monitoring of diabetic patients in a certain hospital as an example, based on medical research and clinical data, various factors that affect the blood sugar levels of diabetic patients, such as eating habits, exercise frequency, medication use, genetic factors, sleep quality, etc., are collected, the collected key factors are screened, duplicate factors are removed, and summarized into a measurable set, and weights are set for each key factor based on industry research and clinical experience, and a risk assessment level is set. The set risk assessment level and weight are reflected to reflect the importance and performance of each key factor in the blood sugar level risk assessment.

[0261] According to another embodiment of the present invention, Figure 2 As shown, a pathological critical value warning system based on a pathological knowledge graph is also provided, and the pathological critical value warning system based on a pathological knowledge graph includes:

[0262] Data acquisition module 1, used to obtain the test index data related to the pathological critical value in the pathological knowledge graph, pre-process the test index data, and obtain the test index characteristic data;

[0263] Feature recognition module 2, used to identify key feature data by analyzing the inspection index feature data obtained by feature classification algorithm;

[0264] Model building module 3, used to build an early warning model based on key feature data, and evaluate the risk level of the inspection indicator through the early warning model;

[0265] Data analysis module 4 is used to analyze the risk level of the inspection indicator through the level analysis algorithm, determine the key factors affecting the risk level of the inspection indicator, and issue early warning reminders based on the key factors;

[0266] The data acquisition module 1 is connected to the model construction module 3 via the feature recognition module 2 , and the model construction module 3 is connected to the data analysis module 4 .

[0267] In summary, with the help of the above technical scheme of the present invention, the present invention ensures the diversity and representativeness of the data sample set by presetting the number of mutation and crossover operations, improves the quality of the feature data sample set by using mutation operations and differential vector calculations, ensures the accurate extraction of key feature data, and determines the clustering affiliation and optimal distance of each test indicator feature data sample through cluster analysis, so as to achieve effective classification of data samples and facilitate further risk assessment and prediction. The present invention constructs an early warning model by using key feature data, and trains and optimizes it to evaluate the risk level of the test indicator, accurately predicts the future risk trend through the construction of cycle and trend indicators, provides forward-looking early warning, evaluates the accuracy of the model by using the mean absolute percentage error, optimizes the model parameters, ensures the accuracy of the prediction results, and achieves high-precision prediction of the risk trend of future test indicators, thereby realizing the efficient construction and optimization of the early warning model, ensuring the reliability of risk assessment, and realizing effective early warning of pathological critical values ​​through fine feature data extraction, efficient feature classification and cluster analysis, dynamic optimization iteration and accurate risk assessment. The present invention ensures the scientificity and comprehensiveness of the evaluation process through detailed data comparison and key factor analysis, combines expert evaluation with fuzzy reasoning to provide authoritative and reliable risk assessment results, ensures timely discovery and response to potential risks through detailed threshold judgment and early warning reminders, and optimizes the pathological critical value management process through comprehensive risk level judgment and key factor analysis, thereby improving medical quality and patient safety.

[0268] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A pathological critical value early warning method based on pathological knowledge graph, characterized in that: The pathological critical value warning method based on the pathological knowledge graph includes the following steps: S1. Obtaining the test index data related to the pathological critical value in the pathological knowledge graph, preprocessing the test index data, and obtaining the test index characteristic data; S2. Using the characteristic classification algorithm to analyze the inspection index characteristic data, identify the key characteristic data; S3. Based on key feature data, build an early warning model and evaluate the risk level of inspection indicators through the early warning model; S4. Analyze the risk level of the inspection indicators through the level analysis algorithm, determine the key factors affecting the risk level of the inspection indicators, and issue early warning reminders based on the key factors; The inspection index characteristic data obtained by analyzing the characteristic classification algorithm and identifying the key characteristic data comprises the following steps: S21, setting a test index characteristic data sample set, and presetting the number of mutation and crossover operations; S22, mutating the test indicator feature data sample set through a mutation operation, calculating the difference vector between each test indicator feature data sample, determining the best feature data sample in the test indicator feature data sample set, and using the best feature data sample as a new mutated test indicator feature data sample; S23, using a feature classification algorithm to perform a new cluster analysis on the test indicator feature data samples, to determine the cluster to which each test indicator feature data sample belongs and the optimal distance between each test indicator feature data sample; S24, analyzing the formed new test index characteristic data sample set, comparing the test index characteristic data sample after mutation crossover with the objective function, and selecting the test index characteristic data sample set with the best objective function value as the new test index characteristic data sample set; S25, judging whether the preset termination condition is met, if so, the evolution is terminated, and the current best test indicator feature data sample set is determined as the key feature data of the test indicator; The method of performing a new cluster analysis on the test index feature data samples by using a feature classification algorithm to determine the cluster to which each test index feature data sample belongs and the optimal distance between each test index feature data sample comprises the following steps: S231, randomly selecting a number of test indicator feature data samples from the test indicator feature data sample set as initial cluster center points; S232, assigning each unassigned test index characteristic data sample to the cluster center point closest to it; S233, randomly selecting a non-cluster center test indicator feature data sample from the test indicator feature data sample set, and recording it as a random test indicator feature data sample; S234, calculating the total cost change caused by replacing the current cluster center point with the random inspection indicator feature data sample, and updating the cluster center set; S235, when the cluster center point no longer changes, determine the final clustering of each test indicator feature data sample, and calculate the optimal distance between each test indicator feature data sample; The process of constructing an early warning model based on key feature data and evaluating the risk level of inspection indicators through the early warning model includes the following steps: S31. Convert the obtained key feature data into a risk assessment cycle; S32, replacing the risk assessment cycle into the cycle index adjustment algorithm, dividing the obtained key feature data into cycle indicators and trend indicators, and constructing and training an early warning model through the cycle indicators and trend indicators to predict the future inspection indicator risk trend; S33. Use the mean absolute percentage error as the evaluation criterion for the accuracy of the early warning model; S34. Based on the model optimization algorithm, adjust and optimize the parameters of the early warning model according to the trend indicators, and find the best parameter configuration; S35. Apply the optimal parameter configuration to the constructed early warning model, use the second-order adaptive coefficient method to predict the trend data, and obtain the risk trend of the inspection indicator at the future moment; S36, converting the obtained inspection indicator risk trend at the future moment into an actual inspection indicator risk assessment, constructing a final early warning model, and evaluating the inspection indicator risk level; The method of adjusting and optimizing the parameters of the early warning model according to the trend indicators and finding the optimal parameter configuration based on the model optimization algorithm includes the following steps: S341, initializing key parameters of the model optimization algorithm; S342, randomly initialize the parameter configuration of the early warning model, calculate the initial risk assessment value of each parameter configuration, and use it as the maximum risk indicator of the parameter; S343. Calculate the relative importance of each risk indicator, and determine the direction of adjusting the early warning model parameters according to the relative importance; S344. Update the early warning model parameter configuration according to the relative importance and trend indicators, and randomly perturb the optimal parameter configuration; S345. Re-evaluate the accuracy and effectiveness of the model based on the updated parameter configuration; S346. If the maximum number of iterations is reached, stop the optimization; otherwise, continue iterating and adjust the parameter configuration; S347. When the optimization process is completed, the best parameter configuration and the optimized early warning model are output.

2. A pathological critical value early warning method based on pathological knowledge graph according to claim 1, characterized in that: The obtaining of the test index data related to the pathological critical value in the pathological knowledge graph, preprocessing the test index data, and obtaining the test index characteristic data comprises the following steps: S11, obtaining duplicate data, missing values ​​and abnormal values ​​of the test index data related to the pathological critical value, and performing denoising, filtering and smoothing on the duplicate data, missing values ​​and abnormal values; S12, connecting the unprocessed data in the inspection index data to generate a new data table, associating different data tables through external key values, generating a complete data table, and obtaining an accurate data set; S13. Extract features from the accurate data set and obtain test indicator feature data.

3. A pathological critical value early warning method based on pathological knowledge graph according to claim 1, characterized in that: The formula for mutating the test index feature data sample set through mutation operation is: ； In the formula, Expressed as Moment, c A sample of characteristic data of the variation test indicators; Represented as parent basis vectors; is the parent difference vector; S Expressed as a scaling factor; t 1. t 2. t 3 represent three different indexes randomly selected from the index of the current test index feature data sample set.

4. A pathological critical value early warning method based on pathological knowledge graph according to claim 1, characterized in that: The formula for updating the early warning model parameter configuration is: ； in, Indicates the current parameter configuration; Represented as the optimal parameter configuration; Expressed as step size factor; It is expressed as the attractiveness coefficient; Represented as a random number.

5. The pathological critical value early warning method based on pathological knowledge graph according to claim 1 is characterized in that: The step of analyzing the risk level of the inspection indicator by using the level analysis algorithm, determining the key factors affecting the risk level of the inspection indicator, and issuing early warning reminders based on the key factors includes the following steps: S41. Obtain risk levels of different inspection indicators and compare them with industry standard risk assessment indicators; S42. Analyze the key factors that affect the risk level of inspection indicators and set the corresponding risk assessment level; S43, using a grade analysis algorithm to train the obtained risk levels of different inspection indicators, calculate the risk score, and determine whether it belongs to a high risk level; S44, normalizing the risk score obtained by the grade analysis algorithm and calculating the fuzzy set; S45. Analytical experts evaluate each key factor at different risk levels to form a risk assessment matrix; S46. Using fuzzy reasoning to calculate the normalized risk score and the risk assessment matrix to obtain a comprehensive risk assessment result; S47. Determine the relative position of the inspection indicators in different risk levels based on the component values ​​of the comprehensive evaluation results and determine the risk level; S48, determining whether the risk level exceeds a set threshold, if so, marking it as high risk, if not, treating it as low or medium risk; S49. Based on the results of the grade analysis algorithm and the advice of analysis experts, determine the key factors that affect the risk level of the inspection indicators and issue warnings based on the key factors.

6. A pathology critical value early warning method based on pathology knowledge graph according to claim 5, characterized in that: The analysis of the key factors affecting the risk level of the inspection indicators and setting the corresponding risk assessment level includes the following steps: S421. Collect various key factors that affect the risk level of inspection indicators based on industry research and inspection data; S422. Screen and classify the key factors that affect the risk level of various inspection indicators collected, remove duplicate factors, and summarize them into a set of measurable key factors; S423. Set weights for each key factor affecting the risk level based on industry research and inspection experience, and set risk assessment levels for each key factor; S424. Reflect the importance and performance of each key factor in the risk assessment of the inspection indicator based on the set risk assessment level.

7. A pathology critical value warning system based on a pathology knowledge graph, used to implement the pathology critical value warning method based on a pathology knowledge graph as described in any one of claims 1 to 6, characterized in that: The pathological critical value warning system based on the pathological knowledge graph includes: A data acquisition module is used to obtain the test index data related to the pathological critical value in the pathological knowledge graph, pre-process the test index data, and obtain the test index characteristic data; A feature recognition module is used to identify key feature data by analyzing the inspection index feature data obtained by the feature classification algorithm; Model building module, used to build early warning models based on key feature data, and evaluate the risk level of inspection indicators through early warning models; The data analysis module is used to analyze the risk level of the inspection indicators through the level analysis algorithm, determine the key factors affecting the risk level of the inspection indicators, and issue early warning reminders based on the key factors; Among them, the data acquisition module is connected with the model construction module through the feature recognition module, and the model construction module is connected with the data analysis module.

Images