A method and system for big data analysis of medical decision-making
Through the classification, feature extraction and rule matching of medical data, combined with logistic regression model, the rationality of auxiliary medical decision-making strategies is verified, and the problems of the accuracy and consistency of auxiliary medical decision-making among doctors are solved, and more efficient medical decision-making is achieved process.
Patent Information
- Application Number
- CN202410981693.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-07-22
AI Technical Summary
When different doctors review the auxiliary medical decisions output by the artificial intelligence model, there are differences in accuracy and consistency, mainly due to differences in experience level and understanding ability.
By obtaining medical data and classifying, extracting features, using knowledge graphs and clinical rules to match rules, generating explanatory data, using logistic regression models to predict future trends, and verifying whether the results are in line with knowledge graphs and clinical rules to determine auxiliary medical decisions Strategy.
It improves the accuracy and consistency of auxiliary medical decisions, reduces the impact of individual differences between doctors, ensures the rationality and transparency of decisions, is suitable for doctors with less experience, and enhances the review efficiency and credibility of model output.
Smart Images

Figure CN118522469B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent medical technology, and in particular to a big data analysis medical decision-making method and system. Background Art
[0002] Big data analysis of medical decision-making generally uses big data technology and analysis methods to process and analyze data in the medical field to assist in medical decision-making and improve the quality and efficiency of medical services. By collecting and integrating large amounts of medical data, such as clinical records, medical images, and genomics data, combined with machine learning and artificial intelligence for analysis and prediction, it helps medical decision-makers better understand the pathogenesis of diseases, the effectiveness of diagnosis and treatment methods, and thus obtain auxiliary medical decision-making strategies.
[0003] In the existing technology, in order to ensure that the analysis and prediction results and auxiliary decision-making suggestions output by artificial intelligence models in the medical field are highly accurate and reliable, the conventional practice is for doctors to conduct a secondary review. The accuracy of this review process is highly dependent on the doctor's professional knowledge, clinical experience and judgment. However, given the differences in experience levels among different doctors, different doctors have different understandings of the output results of the same model, resulting in different review efficiency and accuracy of the same model output results. Summary of the invention
[0004] The problem solved by the present invention is how to improve the accuracy and consistency of auxiliary medical decision-making output by different doctor review models.
[0005] To solve the above problems, the present invention provides a big data analysis medical decision-making method, comprising:
[0006] Acquire medical data of the diagnostic target;
[0007] Classifying the medical data to obtain at least one medical field corresponding to the medical data;
[0008] Obtaining the knowledge graph and clinical rules corresponding to the medical field;
[0009] Performing feature extraction on the medical data to obtain a plurality of data features of the medical data, wherein each of the data features corresponds to one or more of the medical fields;
[0010] According to the data features of the medical data, the relevant features of the diagnosis target are obtained by screening through feature selection technology;
[0011] Performing rule matching on the relevant features according to the knowledge graph of the medical field and the clinical rules corresponding to the relevant features to obtain explanatory data of the relevant features;
[0012] Obtaining the change trend of the relevant features within a preset future time period according to the relevant features through a logistic regression model;
[0013] Verifying whether the change trend conforms to the knowledge graph and the clinical rules in the medical field according to the explanatory data;
[0014] If so, an auxiliary medical decision-making strategy for the diagnostic target is determined based on the change trend.
[0015] Optionally, the data features include: distribution features, time series features, text features, and image features, and the feature extraction of the medical data to obtain multiple data features of the medical data includes:
[0016] Determine whether there are abnormal values in the medical data, and determine whether the medical data is complete;
[0017] When the medical data does not contain the abnormal value and the medical data is complete, dividing the medical data into multiple groups of numerical data, time series data, text data and image data;
[0018] Performing feature statistics on the numerical data to obtain the distribution feature corresponding to each group of numerical data;
[0019] Extracting frequency domain and time domain features from the time series data to obtain the time series features corresponding to each group of the time series data;
[0020] Extract keywords from the text data to obtain the text features corresponding to each group of the text data;
[0021] Image features are extracted from the image data to obtain the image features corresponding to each group of the image data.
[0022] Optionally, the step of obtaining relevant features of the diagnosis target by screening according to the data features of the medical data through a feature selection technique includes:
[0023] Determine the correlation coefficient between the data feature of the medical data and the diagnosis target by using a filter filtering method;
[0024] Determining the relevant features of the diagnostic target according to the correlation coefficient;
[0025] The relevant features are sorted from large to small according to the correlation coefficient to obtain a feature sequence, and then a preset number of the data features are selected in order from the feature sequence as the relevant features.
[0026] Optionally, determining the correlation coefficient between the data feature of the medical data and the diagnosis target by using a filter filtering method includes:
[0027] Normalizing the data features to unify the data features to the same magnitude, thereby obtaining the processed data features;
[0028] Determining the type of the correlation coefficient according to the data type in the data feature, the type of the correlation coefficient including the Pearson correlation coefficient, the Spearman rank correlation coefficient and the point-biserial correlation coefficient;
[0029] The weight of the data feature is adjusted by an adaptive weight allocation technique; and the correlation coefficient between the data feature and the diagnostic target is determined based on the calculation formula corresponding to the adjusted weight, the Pearson correlation coefficient, the Spearman rank correlation coefficient and the point-biserial correlation coefficient.
[0030] Optionally, performing rule matching on the relevant features according to the knowledge graph of the medical field corresponding to the relevant features and the clinical rules to obtain explanatory data of the relevant features includes:
[0031] Performing natural language recognition on the relevant features according to the knowledge graph to obtain an initial keyword mapping table of the relevant features in the knowledge graph; wherein the initial keyword mapping table includes all common expressions of the relevant features and entity mappings of the common expressions in the knowledge graph;
[0032] Supplementing the initial keyword mapping table through similarity matching to obtain a final keyword mapping table;
[0033] Encoding the clinical rules to convert the clinical rules into logical expressions;
[0034] Determining whether the relevant feature complies with the clinical rule through the logical expression, and obtaining the determination result of the relevant feature;
[0035] The explanatory data of the relevant features are generated according to the judgment result and the final keyword mapping table.
[0036] Optionally, obtaining the change trend of the relevant features within a preset future time period according to the relevant features through a logistic regression model includes:
[0037] Obtain multiple sets of forecast scenarios;
[0038] Inputting the relevant features into the logistic regression model to obtain the change probability of the relevant features under each group of the prediction scenarios within the preset future time period;
[0039] According to the change probability, obtaining the probability distribution of the relevant feature in the preset future time period;
[0040] Through the probability distribution, the change trend of the relevant features in each group of the prediction scenarios within a preset future time period is obtained.
[0041] Optionally, verifying whether the change trend conforms to the knowledge graph and the clinical rules in the medical field according to the explanatory data includes:
[0042] According to the change trend, obtain the prediction logic and change description corresponding to the change trend;
[0043] Determining whether the change descriptions are associated according to the final keyword mapping table;
[0044] Whether the prediction logic is consistent is determined according to the logical expression of the clinical rule.
[0045] Optionally, the verifying whether the change trend conforms to the knowledge graph and the clinical rules in the medical field according to the explanatory data further includes:
[0046] When the final keyword mapping table is associated with the change description and the logical expression of the clinical rule is consistent with the prediction logic, the change trend is consistent with the knowledge graph and the clinical rule in the medical field.
[0047] Optionally, the auxiliary medical decision-making strategy for determining the diagnostic target according to the change trend includes:
[0048] Analyze the diagnostic target according to the change trend, and determine an adaptive threshold value of the diagnostic target in the medical field;
[0049] Determining a diagnostic standard of the diagnostic target according to a relationship between the change trend and the adaptive threshold;
[0050] Determine the auxiliary medical decision-making strategy for the diagnostic target according to the diagnostic criteria and the change trend;
[0051] The step of analyzing the diagnostic target according to the change trend and determining the adaptive threshold of the diagnostic target in the medical field includes:
[0052] Acquire diagnostic threshold data corresponding to the medical field, and use the diagnostic threshold data as an initial threshold;
[0053] Inputting the change trend into an adaptive threshold model for optimization, and adjusting the initial threshold according to the change trend;
[0054] When the optimization of the adaptive threshold model is completed, it is determined that the adjustment of the initial threshold is completed, and the initial threshold after the adjustment is completed is used as the adaptive threshold.
[0055] The present invention also provides a big data analysis medical decision-making system, comprising:
[0056] A first data acquisition unit, used to acquire medical data of a diagnosis target;
[0057] A classification unit, configured to classify the medical data to obtain at least one medical field corresponding to the medical data;
[0058] A second data acquisition unit, used to acquire the knowledge graph and clinical rules corresponding to the medical field;
[0059] A feature extraction unit, configured to extract features from the medical data to obtain a plurality of data features of the medical data, wherein each of the data features corresponds to one or more of the medical fields;
[0060] A feature screening unit, configured to screen the data features of the medical data using a feature selection technique to obtain relevant features of the diagnosis target;
[0061] An explanatory data acquisition unit, configured to perform rule matching on the relevant features according to the knowledge graph of the medical field and the clinical rules corresponding to the relevant features, so as to obtain explanatory data of the relevant features;
[0062] A prediction unit, used to obtain a change trend of the relevant features within a preset future time period according to the relevant features through a logistic regression model;
[0063] A judgment unit, used for verifying whether the change trend conforms to the knowledge graph and the clinical rules in the medical field according to the explanatory data;
[0064] A decision-making unit is used to determine, if yes, an auxiliary medical decision-making strategy for the diagnostic target according to the change trend.
[0065] The big data analysis medical decision-making method and system of the present invention first obtains the medical data of the diagnostic target and classifies it to ensure that the data corresponds to the corresponding medical field; and each data feature corresponds to one or more medical fields, which helps to analyze the problem from different angles and increase the comprehensiveness of the decision. Then, for each medical field, the corresponding knowledge graph and clinical rules are obtained, wherein the knowledge graph provides the complex relationship between diseases, symptoms, diagnosis, treatment, etc., and the clinical rules contain the standardized operations and experience summaries in medical practice. The combination of the two provides rich background knowledge and rule support for medical data analysis; standardized knowledge graphs and clinical rules can provide a unified reference framework for medical decision-making, thereby helping to reduce inconsistencies in decision-making due to differences in personal experience of doctors. Then, feature extraction is performed on the medical data, and features closely related to the diagnostic target are screened out through feature selection technology. By focusing on key features, the accuracy of decision-making is further improved. For these related features, rule matching is performed through the knowledge graph and clinical rules in the medical field to generate explanatory data. The explanatory data provides specific and clear explanatory information for auxiliary medical decision-making strategies. Even doctors with low experience levels can understand the logic and basis of decision-making on diagnostic targets, assist doctors in understanding auxiliary medical decision-making strategies, and help doctors confirm the rationality of model output. In addition, the logistic regression model is used to obtain the changing trend of related features in the future time period, which enhances the transparency and credibility of the decision-making process. According to the explanatory data generated by the rules and knowledge graph in the medical field, the changing trend of related features in the future time period is also obtained. Among them, the use of the logistic regression model to predict the changing trend of related features in the future time period can understand possible medical conditions and risks in advance, so as to make more forward-looking decisions. Finally, the generated explanatory data is used to verify whether these changing trends conform to the rules and knowledge graphs in the medical field, ensuring the rationality and accuracy of the decision. If the prediction results are verified, the auxiliary medical decision-making strategy is determined according to the changing trend. In addition, the auxiliary medical decision-making strategy of the present invention has undergone standardized knowledge verification, which is more suitable for the medical field, making it easier for doctors to improve the accuracy of the review and easier to understand, thereby improving the review efficiency of doctors with lower levels, reducing the impact of individual differences among doctors, and further improving the consistency of decisions. At the same time, since the formulation of the auxiliary medical decision-making strategy strictly follows the knowledge graph and clinical rules, even doctors with less experience can review the output of the model more accurately, further improving the consistency of the auxiliary medical decision review output by the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0066] Figure 1 This is a flowchart of a method for medical decision-making based on big data analysis in an embodiment of the present invention;
[0067] Figure 2It is a structural diagram of the big data analysis medical decision-making method system in an embodiment of the present invention. DETAILED DESCRIPTION
[0068] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.
[0069] Combination Figure 1 As shown, the present invention provides a big data analysis medical decision-making method, comprising:
[0070] Obtain medical data of the diagnostic target.
[0071] Specifically, data is generally collected from different medical information systems, including electronic medical records, laboratory test results, imaging data, etc. In a preferred embodiment of the present invention, data crawling technology can be used to obtain information from public resources.
[0072] The medical data is classified to obtain at least one medical field corresponding to the medical data.
[0073] Specifically, the medical data is classified to obtain at least one medical field corresponding to the medical data, and the medical data is assigned to one or more medical fields, such as cardiology, neurology, etc. In a preferred embodiment of the present invention, the medical data set is generally preprocessed first, such as data cleaning and standardization, and then a classification model is trained to classify the data.
[0074] Obtain the knowledge graph and clinical rules corresponding to the medical field.
[0075] Specifically, the knowledge graph and clinical rules corresponding to the medical field are obtained, where the knowledge graph is usually a large graphical database containing medical concepts, clinical rules and their interrelationships. This step extracts information from professional medical knowledge bases, clinical pathways and expert experience to build or update the knowledge graph. The acquisition of clinical rules may involve working with medical experts to codify their professional knowledge into actionable rules.
[0076] Feature extraction is performed on the medical data to obtain multiple data features of the medical data, wherein each of the data features corresponds to one or more medical fields.
[0077] Specifically, in a preferred embodiment of the present invention, natural language processing technology can be used to extract useful information from unstructured medical records, or image processing algorithms can be used to extract features from medical images.
[0078] According to the data features of the medical data, relevant features of the diagnosis target are obtained by screening through feature selection technology.
[0079] Specifically, relevant features of the diagnostic target are screened through feature selection technology, wherein the feature selection technology may be recursive feature elimination, etc., so as to screen out the features most relevant to the diagnostic target from a group of candidate data features.
[0080] The relevant features are rule matched according to the knowledge graph of the medical field and the clinical rules corresponding to the relevant features to obtain explanatory data of the relevant features.
[0081] Specifically, through rule matching, each feature will be associated with one or more specific medical explanations, which come from the matching results of the knowledge graph and clinical rules. The explanatory data may include disease diagnosis, pathological mechanism, treatment recommendations, etc. In addition, the generated explanatory data needs to be further verified to see if it is consistent with current medical knowledge and clinical practice. This step may require the participation of medical experts to ensure the accuracy and practicality of the explanation.
[0082] The logistic regression model is used to obtain the change trend of the relevant features within a preset future time period based on the relevant features.
[0083] Specifically, a logistic regression model is used to analyze the changing trend of the selected features within a preset future time period. Therefore, it is necessary to collect historical data as a training set, and then train and verify the prediction performance of the model based on these data. When the prediction performance of the model meets the requirements, it can be used as a logistic regression model to predict the changing trend.
[0084] Verify whether the change trend is consistent with the knowledge graph and the clinical rules in the medical field based on the explanatory data.
[0085] Specifically, after obtaining explanatory data for specific medical data features, it is necessary to use these data to verify the change trend and compare it with the knowledge graph in the medical field. First, find the paths and entity relationships that match the explanatory data in the knowledge graph, and verify whether the logical relationship between the data features is consistent with the entities in the knowledge graph. For example, whether the change trend of a specific symptom or biochemical indicator is consistent with the development stage of a disease. Use the graph reasoning algorithm to analyze the position and relationship of the explanatory data in the graph, and check whether there is a chain of evidence supporting this data.
[0086] If so, an auxiliary medical decision-making strategy for the diagnostic target is determined based on the change trend.
[0087] Specifically, based on the results of trend analysis and clinical significance assessment, decision-making support strategies are developed, which may involve adjustments to diagnostic tests, treatment options, disease monitoring, and preventive measures. For example, if data analysis indicates that a patient population has a poor response trend to a specific drug, a more effective treatment option may be considered.
[0088] The big data analysis medical decision-making method of the present invention first obtains the medical data of the diagnostic target and classifies it to ensure that the data corresponds to the corresponding medical field; and each data feature corresponds to one or more medical fields, which helps to analyze the problem from different angles and increase the comprehensiveness of the decision. Then, for each medical field, the corresponding knowledge graph and clinical rules are obtained, wherein the knowledge graph provides the complex relationship between diseases, symptoms, diagnosis, treatment, etc., and the clinical rules contain the standardized operations and experience summaries in medical practice. The combination of the two provides rich background knowledge and rule support for medical data analysis; standardized knowledge graphs and clinical rules can provide a unified reference framework for medical decision-making, thereby helping to reduce inconsistencies in decision-making due to differences in personal experience of doctors. Then, feature extraction is performed on the medical data, and features closely related to the diagnostic target are screened out through feature selection technology. By focusing on key features, the accuracy of decision-making is further improved. For these related features, rule matching is performed through the knowledge graph and clinical rules in the medical field to generate explanatory data. The explanatory data provides specific and clear explanatory information for auxiliary medical decision-making strategies. Even doctors with low experience levels can understand the logic and basis of decision-making on diagnostic targets, assist doctors in understanding auxiliary medical decision-making strategies, and help doctors confirm the rationality of model output. In addition, the logistic regression model is used to obtain the changing trend of related features in the future time period, which enhances the transparency and credibility of the decision-making process. According to the explanatory data generated by the rules and knowledge graph in the medical field, the changing trend of related features in the future time period is also obtained. Among them, the use of the logistic regression model to predict the changing trend of related features in the future time period can understand possible medical conditions and risks in advance, so as to make more forward-looking decisions. Finally, the generated explanatory data is used to verify whether these changing trends conform to the rules and knowledge graphs in the medical field, ensuring the rationality and accuracy of the decision. If the prediction results are verified, the auxiliary medical decision-making strategy is determined according to the changing trend. In addition, the auxiliary medical decision-making strategy of the present invention has undergone standardized knowledge verification, which is more suitable for the medical field, making it easier for doctors to improve the accuracy of the review and easier to understand, thereby improving the review efficiency of doctors with lower levels, reducing the impact of individual differences among doctors, and further improving the consistency of decisions. At the same time, since the formulation of the auxiliary medical decision-making strategy strictly follows the knowledge graph and clinical rules, even doctors with less experience can review the output of the model more accurately, further improving the consistency of the auxiliary medical decision review output by the model.
[0089] Optionally, classifying the medical data to obtain at least one medical field corresponding to the medical data includes:
[0090] Determine basic information and test results of the diagnosis target according to the medical data;
[0091] Inputting the basic information and the detection results into a machine learning classification model;
[0092] Through the output of the machine learning classification model, at least one medical field corresponding to each of the detection results is obtained.
[0093] Specifically, first, it is necessary to collect and organize the basic information of the patient's diagnostic target, including age, gender, family history, etc., and results from various medical tests, such as blood analysis, imaging examination, biochemical indicators, etc. These data are used as the input of the machine learning classification model. The task of the model is to analyze the characteristics of the input and then predict the medical field to which these data are most likely to belong. For example, in a specific example of implementing this method, a support vector machine (SVM), a random forest or a deep learning network can be used as a classification model. In another preferred embodiment of the present invention, a deep learning model is used. The input of the model is a numerical representation of the patient's basic information and the numerical or image data of each test result. By learning the patterns in the existing data set, the medical field corresponding to the new patient data is predicted, such as cardiology, neurology or oncology.
[0094] In this embodiment, medical data is classified into relevant fields in an automated manner, which speeds up data processing and improves the accuracy of data classification.
[0095] Optionally, the data features include: distribution features, time series features, text features, and image features, and the feature extraction of the medical data to obtain multiple data features of the medical data includes:
[0096] Determine whether there are abnormal values in the medical data, and determine whether the medical data is complete;
[0097] When the medical data does not contain the abnormal value and the medical data is complete, dividing the medical data into multiple groups of numerical data, time series data, text data and image data;
[0098] Performing feature statistics on the numerical data to obtain the distribution feature corresponding to each group of numerical data;
[0099] Extracting frequency domain and time domain features from the time series data to obtain the time series features corresponding to each group of the time series data;
[0100] Extract keywords from the text data to obtain the text features corresponding to each group of the text data;
[0101] Image features are extracted from the image data to obtain the image features corresponding to each group of the image data.
[0102] Specifically, feature engineering is a crucial step in the process of big data processing and analysis, which involves the process of converting raw data into features that can be directly used by the model. The above content describes a feature extraction method for medical data, which first determines the integrity and outliers of medical data to ensure the quality of data entering the analysis process. This preliminary judgment step is basic but very critical, because data containing outliers or incomplete data may lead to inaccurate analysis results. Then, according to the different types of data, feature extraction is performed separately: for the processing of numerical data, distribution features are obtained by performing feature statistics, which may include calculating the mean, median, variance, and extreme values, etc. These features can help the model understand the distribution of data. For the processing of time series data, frequency domain and time domain feature extraction are used to reflect the change law of data over time, which is particularly important for medical scenarios that need to analyze the change of data over time (such as electrocardiogram analysis). For the processing of text data, through keyword extraction, key medical terms or symptom descriptions can be identified from medical reports, providing a basis for subsequent text analysis and information retrieval. Image data processing and image feature extraction may include edge detection, texture analysis, and morphological analysis, which are crucial for the analysis of medical images such as X-rays. For example, in the diagnosis of heart disease, the use of time series feature extraction technology can identify small changes in heart rhythm from the electrocardiogram.
[0103] In this embodiment, by judging the integrity and outliers of the data, the accuracy and reliability of the analysis are ensured, and misleading conclusions caused by data quality issues are avoided. Secondly, the medical data is grouped and processed according to type, making feature extraction more targeted and effective, and improving the depth and breadth of the analysis. The distribution feature extraction of numerical data helps to reveal the statistical characteristics of the data and provides a comprehensive view of the data distribution for the model. The frequency domain and time domain feature extraction of time series data can capture the dynamic changes of time series and help understand medical data that changes over time, such as physiological signals. Keyword extraction of text data makes it possible to extract useful information from unstructured medical records, enhancing the application capabilities of natural language processing in the medical field. Finally, image feature extraction of image data provides support for medical image analysis by identifying visual information such as edges and textures in images.
[0104] Optionally, the step of obtaining relevant features of the diagnosis target by screening according to the data features of the medical data through a feature selection technique includes:
[0105] Determine the correlation coefficient between the data feature of the medical data and the diagnosis target by using a filter filtering method;
[0106] Determining the relevant features of the diagnostic target according to the correlation coefficient;
[0107] The relevant features are sorted from large to small according to the correlation coefficient to obtain a feature sequence, and then a preset number of the data features are selected in order from the feature sequence as the relevant features.
[0108] Specifically, screening and obtaining features related to the diagnostic target through feature selection technology is a key step to improve model performance and interpretability. Among them, the Filter filtering method is a commonly used feature selection method, which is based on statistical principles and directly evaluates the importance of features by calculating the correlation coefficient between features and target variables without relying on any prediction model. The advantage of this method is that it has high computational efficiency, is easy to implement, and is suitable for processing large-scale data sets. In the implementation process, the correlation coefficient between each data feature and the diagnostic target (such as the presence or absence of the disease, the severity of the disease, etc.) is first calculated. These correlation coefficients may be calculated using the Pearson correlation coefficient, the Spearman rank correlation coefficient, or other statistical methods, depending on the type and distribution characteristics of the data. The calculated correlation coefficient reflects the strength of the correlation between each feature and the diagnostic target. Subsequently, all features are sorted according to the size of their correlation with the diagnostic target according to these correlation coefficients to form a feature sequence. From this sequence, a preset number of the most relevant features can be selected in order as the final set of relevant features for subsequent analysis or model construction. In a preferred embodiment of the present invention, in a study on the prediction of cardiovascular disease, clinical indicator data of patients are collected, including dozens of features such as blood pressure, cholesterol level, and body mass index (BMI). By calculating the correlation coefficients between these features and the occurrence of cardiovascular diseases, it may be found that cholesterol levels and BMI are highly correlated with cardiovascular diseases. Subsequently, the researchers can select these features with the highest correlation coefficients for in-depth analysis or to build a prediction model. The feature selection method based on the Filter filtering method can effectively identify the most important factors for the prediction of cardiovascular diseases from a large number of clinical indicators, providing a scientific basis for the formulation of targeted preventive measures and treatment plans.
[0109] In this embodiment, by selecting features that are highly relevant to the diagnostic target, the interference of irrelevant or redundant information on the model can be reduced, thereby improving the accuracy and generalization ability of the model. The feature set selected by the Filter method has a clear quantification of the correlation between these features and the diagnostic target, which helps doctors understand the decision logic of the model. The correlation between these features and the diagnostic target has been clearly quantified. Secondly, feature selection helps to reduce the complexity of the model, reduce the consumption of computing resources, and speed up the training of the model. In addition, through feature selection, the interpretability of the model can be improved, because the model only relies on a few important features, which makes the decision-making process of the model more transparent and easier for medical experts to understand and trust.
[0110] Optionally, determining the correlation coefficient between the data feature of the medical data and the diagnosis target by using a filter filtering method includes:
[0111] Normalizing the data features to unify the data features to the same magnitude, thereby obtaining the processed data features;
[0112] Determining the type of the correlation coefficient according to the data type in the data feature, the type of the correlation coefficient including the Pearson correlation coefficient, the Spearman rank correlation coefficient and the point-biserial correlation coefficient;
[0113] The weight of the data feature is adjusted by an adaptive weight allocation technique; and the correlation coefficient between the data feature and the diagnostic target is determined based on the calculation formula corresponding to the adjusted weight, the Pearson correlation coefficient, the Spearman rank correlation coefficient and the point-biserial correlation coefficient.
[0114] Specifically, when determining the correlation coefficient between medical data features and diagnostic targets through the Filter filtering method, the data features need to be normalized first. Normalization is a common data preprocessing technique that scales data features to a specified range, usually [0, 1] or [-1, 1], through linear transformation to eliminate the influence of different magnitudes or units on the calculation of correlation coefficients. The normalized data features are comparable, which helps to accurately calculate the correlation between the features and the diagnostic targets. Next, select the appropriate correlation coefficient type according to the data type of the data feature. The Pearson correlation coefficient is applicable to continuous variables, which measures the degree of linear correlation between two variables; the Spearman rank correlation coefficient is applicable to ordered variables or non-normally distributed continuous variables, which calculates the correlation based on the rank of the variable rather than the original value; the point-biserial correlation coefficient is applicable to the case where one variable is a binary variable (such as gender) and the other variable is a continuous variable. When calculating the correlation coefficient, an adaptive weight allocation mechanism is introduced to design a weight adjustment strategy based on the volatility, conflict (i.e., correlation with other features) or other statistical characteristics of the data features. For example, the CRITIC weight method can be used to calculate the weight based on data volatility and conflict. The initial weight of each feature is calculated using the CRITIC weight method and other methods. The weight is determined by the standard deviation (volatility) and correlation coefficient (conflict) of the feature. The larger the standard deviation, the higher the weight; the larger the correlation coefficient, the lower the weight. Then, the weight is dynamically adjusted according to the iterative progress of the algorithm or the feedback of the model. For example, the whale optimization algorithm can be used, in which the weight consists of two parts, one part adjusts the number of population iterations, and the other part adjusts the local optimum. The weight value will change adaptively according to the change of the population position. The weight is calculated by dynamically adjusting the correlation coefficient of different features according to the importance of the feature to highlight the more influential features. And according to the specific data type and distribution characteristics, select a suitable calculation formula to determine the correlation coefficient between the data feature and the diagnostic target. According to the adjusted weight and the data type of the data feature, select a suitable correlation coefficient calculation formula, such as the Pearson correlation coefficient, the Spearman rank correlation coefficient or the point-biserial correlation coefficient, to determine the correlation between the data feature and the diagnostic target. So as to evaluate the importance of the feature. In a preferred embodiment of the present invention, in the diabetes prediction model, researchers may collect multiple data features including age, weight, blood sugar level, etc. After normalization, the correlation between these features and the diabetes diagnosis results is calculated using the Pearson correlation coefficient. The results may show that the correlation coefficient between blood sugar level and diabetes is the highest, indicating that blood sugar level is a key feature for predicting diabetes. Based on this finding, more attention can be paid to blood sugar monitoring, so that intervention measures can be taken at an early stage to improve the treatment effect.
[0115] In this embodiment, the normalization process ensures that data features of different magnitudes and units of measurement can be compared and analyzed on the same scale, which helps to avoid the deviation caused by inconsistent dimensions when calculating the correlation coefficient. Secondly, the adaptive weight allocation technology can dynamically adjust the weights according to the statistical characteristics of the data features, which not only enhances the sensitivity of the model to important features, but also reduces the interference of noise and irrelevant features.
[0116] Optionally, performing rule matching on the relevant features according to the knowledge graph of the medical field corresponding to the relevant features and the clinical rules to obtain explanatory data of the relevant features includes:
[0117] Performing natural language recognition on the relevant features according to the knowledge graph to obtain an initial keyword mapping table of the relevant features in the knowledge graph; wherein the initial keyword mapping table includes all common expressions of the relevant features and entity mappings of the common expressions in the knowledge graph;
[0118] Supplementing the initial keyword mapping table through similarity matching to obtain a final keyword mapping table;
[0119] Encoding the clinical rules to convert the clinical rules into logical expressions;
[0120] Determining whether the relevant feature complies with the clinical rule through the logical expression, and obtaining the determination result of the relevant feature;
[0121] The explanatory data of the relevant features are generated according to the judgment result and the final keyword mapping table.
[0122] Specifically, through natural language recognition technology, the relevant features are initially mapped to the entities in the knowledge graph to form an initial keyword mapping table. This mapping table includes all common expressions of the relevant features and their corresponding entities in the knowledge graph, and establishes the basic connection between the data features and the entities in the knowledge graph. Then, through similarity matching, this mapping table is supplemented and improved, and the precise association between the relevant features and the entities in the knowledge graph is finally determined to form the final keyword mapping table. At the same time, the clinical rules are encoded and converted into operable logical expressions, which allows the use of computational methods to directly apply these clinical rules to determine whether the relevant features meet the standard procedures or treatment guidelines in the medical field. Combined with the judgment results of the logical expression and the final keyword mapping table, the system can generate explanatory data that explains how each feature is associated with the knowledge system and clinical practice in the medical field. In a preferred embodiment of the present invention, when performing blood sugar control analysis, if the relevant features (such as records in a blood sugar diary) match the entities in the knowledge graph (such as "blood sugar level") and clinical rules (such as target values for blood sugar control), specific and personalized diet and medication recommendations can be generated.
[0123] In this embodiment, by combining natural language recognition technology and knowledge graph to match the key features in medical data, the accuracy and personalization level of clinical decision-making are improved. First, by using natural language recognition technology, keywords that match the medical knowledge graph can be extracted from unstructured text to form an initial keyword mapping table, which provides a basis for the semantic understanding of medical data. Subsequently, the application of similarity matching further enriches and refines the mapping table, ensuring the accurate correspondence between the features and the entities in the knowledge graph. In addition, the clinical rules are encoded as logical expressions so that these rules can be applied to the evaluation of relevant features in a computational manner, thereby automatically determining whether the features meet the established clinical standards; and the logical expressions are used to evaluate the relevant features to determine whether they meet the clinical rules. This automatic judgment process improves the accuracy of inexperienced doctors in review. Finally, the judgment results and the final keyword mapping table are combined to generate explanatory data, which not only contains the evaluation results of the features, but also provides direct links to the entities in the knowledge graph, enhancing the transparency and comprehensibility of the data. At the same time, because the explanatory data provides a clear logical chain and intuitive reference, doctors can identify key information more quickly during review, thereby improving the efficiency of the review.
[0124] Optionally, obtaining the change trend of the relevant features within a preset future time period according to the relevant features through a logistic regression model includes:
[0125] Obtain multiple sets of forecast scenarios;
[0126] Inputting the relevant features into the logistic regression model to obtain the change probability of the relevant features under each group of the prediction scenarios within the preset future time period;
[0127] According to the change probability, obtaining the probability distribution of the relevant feature in the preset future time period;
[0128] Through the probability distribution, the change trend of the relevant features in each group of the prediction scenarios within a preset future time period is obtained.
[0129] Specifically, in order to simulate different future scenarios, it is necessary to set multiple groups of prediction scenarios, which may be based on different assumptions, such as different treatment plans, environmental changes or individual differences. Then, the selected relevant features are used as input and predicted by the logistic regression model. The logistic regression model is a statistical model widely used in classification and prediction problems, which can calculate the probability of an event based on the input features. In this process, the model outputs the probability of change of the relevant features in the future time period for each group of prediction scenarios. These probabilities reflect the degree of change of the relevant features in a specific scenario. By summarizing the change probabilities under all scenarios, a probability distribution can be obtained, which describes the possible range and probability of change of the relevant features in the future time period. Finally, based on this probability distribution, the specific change trend of the relevant features under each group of prediction scenarios can be inferred to provide quantitative support for decision-making. In a preferred embodiment of the present invention, it is assumed that there are three prediction scenarios: scenario A is normal diet and exercise, scenario B is to increase the amount of exercise, and scenario C is to adjust the diet structure. By inputting the relevant characteristics of the patient (such as age, weight, current blood sugar level, etc.), the model can predict the probability distribution of blood sugar levels in the next month under each scenario. According to these prediction results, combined with actual conditions, the most appropriate management strategy can be selected. If the model predicts that the probability of a drop in blood sugar levels is highest under scenario B, it may be recommended to increase exercise to control blood sugar. This data-driven prediction method not only improves the scientificity and accuracy of medical decision-making.
[0130] In this embodiment, by setting multiple sets of prediction scenarios, simulating the impact of different external conditions or intervention measures on the results, the adaptability and flexibility of the model are increased. The probability of change output by the logistic regression model provides a quantitative evaluation basis for each scenario. Secondly, the acquisition of probability distribution reveals the uncertainty and possible range of feature changes. Finally, the change trend based on probability distribution provides decision makers with specific guidance under different scenarios, helping them understand the potential consequences of different choices and choose the best course of action.
[0131] Optionally, verifying whether the change trend conforms to the knowledge graph and the clinical rules in the medical field according to the explanatory data includes:
[0132] According to the change trend, obtain the prediction logic and change description corresponding to the change trend;
[0133] Determining whether the change descriptions are associated according to the final keyword mapping table;
[0134] Whether the prediction logic is consistent is determined according to the logical expression of the clinical rule.
[0135] Specifically, first, it is necessary to extract the prediction logic and change description from the change trend. These descriptions should clearly express the possible changes of the features in the future and the logic behind them. The prediction logic refers to the causal relationship or influencing factors that lead to the change trend, while the change description is an intuitive expression of these changes. Then, the final keyword mapping table is used to determine whether the change description is associated with the entity in the knowledge graph. This step ensures the consistency of the change description with medical knowledge, that is, whether the terms and concepts in the change description are correctly mapped to the medical entities in the knowledge graph. At the same time, the logical expression of the clinical rules is used to verify whether the prediction logic is consistent with the existing clinical guidelines and treatment principles. This involves comparing the prediction logic with the clinical rules to ensure that the predicted change trend is medically reasonable and feasible. In a preferred embodiment of the present invention, when predicting the future heart rate change trend of heart disease, the prediction logic may be "with the improvement of drug treatment and lifestyle, the heart rate will gradually decrease." The change description may be "the heart rate decreases from the current 80 beats per minute to 70 beats per minute." Through the final keyword mapping table, it can be confirmed that the terms such as "drug treatment", "lifestyle improvement" and "heart rate reduction" are associated with the entities in the knowledge graph. At the same time, through the logical expression of clinical rules, it is possible to verify whether this change trend conforms to the standard process of heart disease treatment. If the verification results are consistent, the medical correctness of the analysis results is ensured.
[0136] In this embodiment, the prediction logic and change description extracted from the change trend provide clear causal relationships and intuitive expressions for medical decision-making, which helps doctors understand and trust the prediction results of the model. The final keyword mapping table is used to verify the relevance of the change description, ensure the correct use of medical terms and concepts, and strengthen the connection between the change description and medical knowledge. The prediction logic is checked through the logical expression of clinical rules to ensure that the prediction results are always consistent with current medical practices and standard treatment guidelines.
[0137] Optionally, the verifying whether the change trend conforms to the knowledge graph and the clinical rules in the medical field according to the explanatory data further includes:
[0138] When the final keyword mapping table is associated with the change description and the logical expression of the clinical rule is consistent with the prediction logic, the change trend is consistent with the knowledge graph and the clinical rule in the medical field.
[0139] Specifically, in medical data analysis, verifying whether the predicted change trend is consistent with the knowledge graph and clinical rules in the medical field is a key link to ensure the clinical applicability of the analysis results. The above content describes a judgment condition, that is, when the final keyword mapping table confirms that the change description is associated with the entity in the knowledge graph, and the logical expression of the clinical rule verifies that the prediction logic is consistent with the existing clinical guidelines and treatment principles, it can be considered that the predicted change trend is consistent with the knowledge graph and clinical rules in the medical field. This judgment condition involves two key verification steps: first, the final keyword mapping table is used to ensure that the terms and concepts in the change description are correctly mapped to the medical entities in the knowledge graph, which ensures the medical accuracy of the change description; second, the logical expression of the clinical rule is used to verify whether the prediction logic is consistent with clinical practice, which ensures that the predicted change trend is medically reasonable and feasible.
[0140] In this embodiment, the verification process of the keyword mapping table is used to ensure that the terms of the change description accurately correspond to the entities in the medical knowledge graph, which improves the professionalism and accuracy of the description and enhances the depth of understanding of the change trend. The consistency verification of the clinical rule logic expression provides support for medical practice for the prediction logic, ensuring that the prediction of the change trend is consistent with the existing clinical guidelines and treatment principles, thereby improving the clinical feasibility of the prediction results. Through precise data analysis and verification, the present invention provides patients with more customized treatment plans and improves the overall quality and efficiency of medical services.
[0141] Optionally, the auxiliary medical decision-making strategy for determining the diagnostic target according to the change trend includes:
[0142] Analyze the diagnostic target according to the change trend, and determine an adaptive threshold value of the diagnostic target in the medical field;
[0143] Determining a diagnostic standard of the diagnostic target according to a relationship between the change trend and the adaptive threshold;
[0144] Determine the auxiliary medical decision-making strategy for the diagnostic target according to the diagnostic criteria and the change trend;
[0145] The step of analyzing the diagnostic target according to the change trend and determining the adaptive threshold of the diagnostic target in the medical field includes:
[0146] Acquire diagnostic threshold data corresponding to the medical field, and use the diagnostic threshold data as an initial threshold;
[0147] Inputting the change trend into an adaptive threshold model for optimization, and adjusting the initial threshold according to the change trend;
[0148] When the optimization of the adaptive threshold model is completed, it is determined that the adjustment of the initial threshold is completed, and the initial threshold after the adjustment is completed is used as the adaptive threshold.
[0149] Specifically, first, collect relevant medical data of the diagnostic target from the medical information system, such as heart rate, blood pressure and other indicators. Use historical data to determine the standard diagnostic threshold in the medical field. For example, the normal range of heart rate may be set at 60-100 beats / minute. Apply statistical methods such as time series analysis to identify the changing trend of data indicators, including calculating moving averages, detecting trend lines, or using advanced models such as ARIMA to predict future values. Secondly, use an adaptive threshold model to dynamically adjust the initial threshold according to the changing trend of the input, where the model may use machine learning techniques such as support vector machines (SVM) or neural networks, which can learn the pattern of data and predict the optimal adjustment of the threshold. The changing trend is input into the adaptive threshold model and optimized. The model automatically adjusts the threshold to better reflect the current health status of the patient by learning the distribution and trend of the data. According to the output of the model, the initial threshold is adjusted. For example, if the upward trend of the heart rate indicates that the average value is increasing, the model may adjust the upper limit of the heart rate to 105 beats / minute. Use the adjusted adaptive thresholds to update the diagnostic criteria to ensure that they can reflect the latest medical data and trends. Based on the updated diagnostic criteria and the changing trends of continuous monitoring, formulate auxiliary medical decision-making strategies, which may involve adjusting treatment plans, recommending further examinations, or changing the frequency of patient monitoring. Implement a feedback mechanism to use clinical results and patient feedback to further optimize the model to ensure that the adaptive threshold model can continue to improve and adapt to new data. Record the threshold adjustment process and decision reasons in detail to facilitate audit and future analysis.
[0150] In this embodiment, by analyzing the diagnostic target according to the changing trend and determining the adaptive threshold of the diagnostic target in the medical field, the medical decision-making process becomes more flexible and accurate, and can adapt to the individual differences of patients and the progression of chronic diseases, thereby providing patients with more personalized treatment plans.
[0151] Combination Figure 2 As shown, the present invention also provides a big data analysis medical decision-making system, comprising:
[0152] A first data acquisition unit, used to acquire medical data of a diagnosis target;
[0153] A classification unit, configured to classify the medical data to obtain at least one medical field corresponding to the medical data;
[0154] A second data acquisition unit, used to acquire the knowledge graph and clinical rules corresponding to the medical field;
[0155] A feature extraction unit, configured to extract features from the medical data to obtain a plurality of data features of the medical data, wherein each of the data features corresponds to one or more of the medical fields;
[0156] A feature screening unit, configured to screen the data features of the medical data using a feature selection technique to obtain relevant features of the diagnosis target;
[0157] An explanatory data acquisition unit, configured to perform rule matching on the relevant features according to the knowledge graph of the medical field and the clinical rules corresponding to the relevant features, so as to obtain explanatory data of the relevant features;
[0158] A prediction unit, used to obtain a change trend of the relevant features within a preset future time period according to the relevant features through a logistic regression model;
[0159] A judgment unit, used for verifying whether the change trend conforms to the knowledge graph and the clinical rules in the medical field according to the explanatory data;
[0160] A decision-making unit is used to determine, if yes, an auxiliary medical decision-making strategy for the diagnostic target according to the change trend.
[0161] The big data analysis medical decision-making system of the present invention first obtains the medical data of the diagnostic target and classifies it to ensure that the data corresponds to the corresponding medical field; and each data feature corresponds to one or more medical fields, which helps to analyze the problem from different angles and increase the comprehensiveness of the decision. Then, for each medical field, the corresponding knowledge graph and clinical rules are obtained, wherein the knowledge graph provides the complex relationship between diseases, symptoms, diagnosis, treatment, etc., and the clinical rules contain the standardized operations and experience summaries in medical practice. The combination of the two provides rich background knowledge and rule support for medical data analysis; standardized knowledge graphs and clinical rules can provide a unified reference framework for medical decision-making, thereby helping to reduce inconsistencies in decision-making due to differences in personal experience of doctors. Then, feature extraction is performed on the medical data, and features closely related to the diagnostic target are screened out through feature selection technology. By focusing on key features, the accuracy of decision-making is further improved. For these related features, rule matching is performed through the knowledge graph and clinical rules in the medical field to generate explanatory data. The explanatory data provides specific and clear explanatory information for auxiliary medical decision-making strategies. Even doctors with low experience levels can understand the logic and basis of decision-making on diagnostic targets, assist doctors in understanding auxiliary medical decision-making strategies, and help doctors confirm the rationality of model output. In addition, the logistic regression model is used to obtain the changing trend of related features in the future time period, which enhances the transparency and credibility of the decision-making process. According to the explanatory data generated by the rules and knowledge graph in the medical field, the changing trend of related features in the future time period is also obtained. Among them, the use of the logistic regression model to predict the changing trend of related features in the future time period can understand possible medical conditions and risks in advance, so as to make more forward-looking decisions. Finally, the generated explanatory data is used to verify whether these changing trends conform to the rules and knowledge graphs in the medical field, ensuring the rationality and accuracy of the decision. If the prediction results are verified, the auxiliary medical decision-making strategy is determined according to the changing trend. In addition, the auxiliary medical decision-making strategy of the present invention has undergone standardized knowledge verification, which is more suitable for the medical field, making it easier for doctors to improve the accuracy of the review and easier to understand, thereby improving the review efficiency of doctors with lower levels, reducing the impact of individual differences among doctors, and further improving the consistency of decisions. At the same time, since the formulation of the auxiliary medical decision-making strategy strictly follows the knowledge graph and clinical rules, even doctors with less experience can review the output of the model more accurately, further improving the consistency of the auxiliary medical decision review output by the model.
[0162] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0163] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.
Claims
1. A big data analysis method for medical decision-making, characterized in that: include: Acquire medical data of the diagnostic target; Classifying the medical data to obtain at least one medical field corresponding to the medical data; Obtaining the knowledge graph and clinical rules corresponding to the medical field; Performing feature extraction on the medical data to obtain a plurality of data features of the medical data, wherein each of the data features corresponds to one or more of the medical fields; According to the data features of the medical data, the relevant features of the diagnosis target are obtained by screening through feature selection technology; According to the knowledge graph of the medical field corresponding to the relevant features and the clinical rules, the relevant features are matched with rules to obtain explanatory data of the relevant features; specifically including: according to the knowledge graph, natural language recognition is performed on the relevant features to obtain an initial keyword mapping table of the relevant features in the knowledge graph; wherein the initial keyword mapping table includes all common expressions of the relevant features and entity mappings of the common expressions in the knowledge graph; Supplementing the initial keyword mapping table through similarity matching to obtain a final keyword mapping table; Encoding the clinical rules to convert the clinical rules into logical expressions; Determining whether the relevant feature complies with the clinical rule through the logical expression, and obtaining the determination result of the relevant feature; generating the explanatory data of the relevant features according to the judgment result and the final keyword mapping table; Obtaining the change trend of the relevant features within a preset future time period according to the relevant features through a logistic regression model; specifically including: obtaining multiple groups of prediction scenarios; Inputting the relevant features into the logistic regression model to obtain the change probability of the relevant features under each group of the prediction scenarios within the preset future time period; According to the change probability, obtaining the probability distribution of the relevant feature in the preset future time period; Obtaining the change trend of the relevant features under each group of the prediction scenarios within a preset future time period through the probability distribution; Verifying whether the change trend conforms to the knowledge graph and the clinical rules in the medical field according to the explanatory data; specifically including: obtaining the prediction logic and change description corresponding to the change trend according to the change trend; Determining whether the change descriptions are associated according to the final keyword mapping table; Determining whether the prediction logic is consistent according to the logical expression of the clinical rule; When the final keyword mapping table is associated with the change description and the logical expression of the clinical rule is consistent with the prediction logic, the change trend conforms to the knowledge graph and the clinical rule in the medical field; If so, an auxiliary medical decision-making strategy for the diagnostic target is determined based on the change trend.
2. The big data analysis medical decision-making method according to claim 1, characterized in that: The data features include: distribution features, time series features, text features and image features. The feature extraction of the medical data to obtain multiple data features of the medical data includes: Determine whether there are abnormal values in the medical data, and determine whether the medical data is complete; When the medical data does not contain the abnormal value and the medical data is complete, dividing the medical data into multiple groups of numerical data, time series data, text data and image data; Performing feature statistics on the numerical data to obtain the distribution feature corresponding to each group of numerical data; Extracting frequency domain and time domain features from the time series data to obtain the time series features corresponding to each group of the time series data; Extract keywords from the text data to obtain the text features corresponding to each group of the text data; Image features are extracted from the image data to obtain the image features corresponding to each group of the image data.
3. The big data analysis medical decision-making method according to claim 1, characterized in that: The step of obtaining relevant features of the diagnosis target by screening according to the data features of the medical data through feature selection technology includes: Determine the correlation coefficient between the data feature of the medical data and the diagnosis target by using a filter filtering method; Determining the relevant features of the diagnostic target according to the correlation coefficient; The relevant features are sorted from large to small according to the correlation coefficient to obtain a feature sequence, and then a preset number of the data features are selected in order from the feature sequence as the relevant features.
4. The big data analysis medical decision-making method according to claim 3 is characterized in that: The determining the correlation coefficient between the data feature of the medical data and the diagnosis target by using a filter filtering method includes: Normalizing the data features to unify the data features to the same magnitude, thereby obtaining the processed data features; Determining the type of the correlation coefficient according to the data type in the data feature, the type of the correlation coefficient including the Pearson correlation coefficient, the Spearman rank correlation coefficient and the point-biserial correlation coefficient; The weight of the data feature is adjusted by an adaptive weight allocation technique; and the correlation coefficient between the data feature and the diagnostic target is determined based on the calculation formula corresponding to the adjusted weight, the Pearson correlation coefficient, the Spearman rank correlation coefficient and the point-biserial correlation coefficient.
5. The big data analysis medical decision-making method according to claim 1, characterized in that: The auxiliary medical decision-making strategy for determining the diagnostic target according to the change trend includes: Analyze the diagnostic target according to the change trend, and determine an adaptive threshold value of the diagnostic target in the medical field; Determining a diagnostic standard of the diagnostic target according to a relationship between the change trend and the adaptive threshold; Determine the auxiliary medical decision-making strategy for the diagnostic target according to the diagnostic criteria and the change trend; The step of analyzing the diagnostic target according to the change trend and determining the adaptive threshold of the diagnostic target in the medical field includes: Acquire diagnostic threshold data corresponding to the medical field, and use the diagnostic threshold data as an initial threshold; Inputting the change trend into an adaptive threshold model for optimization, and adjusting the initial threshold according to the change trend; When the optimization of the adaptive threshold model is completed, it is determined that the adjustment of the initial threshold is completed, and the initial threshold after the adjustment is completed is used as the adaptive threshold.
6. A big data analysis medical decision-making system, characterized in that: include: A first data acquisition unit, used to acquire medical data of a diagnosis target; A classification unit, configured to classify the medical data to obtain at least one medical field corresponding to the medical data; A second data acquisition unit, used to acquire the knowledge graph and clinical rules corresponding to the medical field; A feature extraction unit, configured to extract features from the medical data to obtain a plurality of data features of the medical data, wherein each of the data features corresponds to one or more of the medical fields; A feature screening unit, configured to screen the data features of the medical data using a feature selection technique to obtain relevant features of the diagnosis target; An explanatory data acquisition unit is used to perform rule matching on the relevant features according to the knowledge graph of the medical field corresponding to the relevant features and the clinical rules to obtain explanatory data of the relevant features; specifically comprising: performing natural language recognition on the relevant features according to the knowledge graph to obtain an initial keyword mapping table of the relevant features in the knowledge graph; wherein the initial keyword mapping table includes all common expressions of the relevant features and entity mappings of the common expressions in the knowledge graph; Supplementing the initial keyword mapping table through similarity matching to obtain a final keyword mapping table; Encoding the clinical rules to convert the clinical rules into logical expressions; Determining whether the relevant feature complies with the clinical rule through the logical expression, and obtaining the determination result of the relevant feature; generating the explanatory data of the relevant features according to the judgment result and the final keyword mapping table; A prediction unit is used to obtain the change trend of the relevant features within a preset future time period according to the relevant features through a logistic regression model; specifically comprising: obtaining multiple groups of prediction scenarios; Inputting the relevant features into the logistic regression model to obtain the change probability of the relevant features under each group of the prediction scenarios within the preset future time period; According to the change probability, obtaining the probability distribution of the relevant feature in the preset future time period; Obtaining the change trend of the relevant features under each group of the prediction scenarios within a preset future time period through the probability distribution; A judgment unit, used to verify whether the change trend conforms to the knowledge graph and the clinical rules in the medical field according to the explanatory data; specifically comprising: obtaining the prediction logic and change description corresponding to the change trend according to the change trend; Determining whether the change descriptions are associated according to the final keyword mapping table; Determining whether the prediction logic is consistent according to the logical expression of the clinical rule; When the final keyword mapping table is associated with the change description and the logical expression of the clinical rule is consistent with the prediction logic, the change trend conforms to the knowledge graph and the clinical rule in the medical field; A decision-making unit is used to determine, if yes, an auxiliary medical decision-making strategy for the diagnostic target according to the change trend.
