Medical data supervision method, platform and medical server

Abstract

The invention provides a medical data supervision method and platform and a medical server, and the method comprises the steps: obtaining a medical data set, and carrying out the clustering processing of the medical data set through employing a clustering algorithm, and obtaining a plurality of data subsets with similar features; an association rule mining algorithm is adopted to analyze internal relations among different types of data elements in the structured data subsets, and association features having important influences on grading decision making are obtained; according to a preset medical data grading evaluation index system, constructing a feature set comprehensively covering various data elements in the medical data set from multiple dimensions of importance, sensitivity and application risk of the data; and carrying out feature selection on the feature set by adopting a feature selection algorithm, and balancing the comprehensiveness and pertinence of the feature set by taking a grading index as an optimization target to obtain a feature subset with remarkable influence and low correlation.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a medical data supervision method, platform and medical server. Background Art

[0002] When the medical data supervision platform manages medical data in a hierarchical manner, it faces the technical problem that the data features are complex and diverse and difficult to fully describe. Medical data covers various types such as patients' personal information, medical records, test results, and imaging data. There are significant differences in the attributes, sources, application scenarios, and sensitivity of different types of data. Taking patients' personal information as an example, ID number, contact information, etc. are sensitive data, while gender and age are relatively insensitive; the chief complaint, current medical history, physical examination, etc. in medical records are mainly text, while medical orders and prescriptions are mostly structured data. To accurately evaluate the importance and sensitivity of various types of medical data, it is necessary to deeply analyze their semantic features and internal connections, but the feature engineering of massive medical data faces many technical problems: how to extract key elements from patient information and convert unstructured text (such as medical records) into structured and computable feature vectors; how to characterize the correlation features between different types of data and discover contextual information that has an important impact on hierarchical decisions; how to balance comprehensiveness and pertinence, highlight sensitive features while covering various data elements and avoiding redundancy. Solving these problems requires the comprehensive application of medical knowledge and data science technology, exploring new methods for medical data feature engineering, and providing more accurate and comprehensive data support for the hierarchical management of medical data supervision. Summary of the invention

[0003] The present invention provides a medical data supervision method, which mainly includes:

[0004] Obtain a medical data set, use a clustering algorithm to cluster the medical data set, and obtain multiple data subsets with similar features; use natural language processing technology to extract key semantic elements of unstructured text data in the data subset, and achieve structured representation through named entity recognition and part-of-speech tagging methods to obtain structured data subsets; use association rule mining algorithms to analyze the intrinsic connections between different types of data elements in the structured data subset, and obtain associated features that have an important impact on grading decisions; according to the preset medical data grading evaluation indicator system, from the dimensions of data importance, sensitivity and application risk, construct a feature set that comprehensively covers all types of data elements in the medical data set; use a feature selection algorithm to select features from the feature set, take the grading index as the optimization goal, balance the comprehensiveness and pertinence of the feature set, and obtain feature subsets with significant impact and low correlation; convert the feature subset into a structured feature vector and input it into a preset classification model. The classification model is obtained by training the support vector machine algorithm through labeled samples to obtain an intelligent decision-making model for medical data grading management; use the intelligent decision-making model to make grading decisions on the medical data set, and formulate protection measures for data of different levels in combination with laws, regulations and industry standards for medical data security and privacy protection.

[0005] The present invention also provides a medical data supervision platform, which mainly includes: a data acquisition module, which is used to acquire a medical data set, and cluster the medical data set using a clustering algorithm to obtain multiple data subsets with similar characteristics; a clustering processing module, which is used to extract key semantic elements of unstructured text data in the data subset using natural language processing technology, and achieve structured representation through named entity recognition and part-of-speech tagging methods to obtain structured data subsets; a semantic extraction module, which is used to analyze the intrinsic connections between different types of data elements in the structured data subset using an association rule mining algorithm to obtain association features that have an important impact on hierarchical decision-making; an association analysis module, which is used to analyze the importance, sensitivity, and other aspects of the data according to a preset medical data hierarchical evaluation index system. The module is used to construct a feature set that comprehensively covers all types of data elements in the medical data set based on the degree and application risk dimensions; the feature construction module is used to select features from the feature set using a feature selection algorithm, taking the grading index as the optimization goal, balancing the comprehensiveness and pertinence of the feature set, and obtaining a feature subset with significant impact and low correlation; the feature selection module is used to convert the feature subset into a structured feature vector, which is input into a preset classification model. The classification model is obtained by training the support vector machine algorithm through labeled samples to obtain an intelligent decision-making model for hierarchical management of medical data; the decision generation module is used to make hierarchical decisions on the medical data set using an intelligent decision-making model, and formulate protection measures for data of different levels in combination with the laws, regulations and industry standards on medical data security and privacy protection.

[0006] The present invention also provides a medical data supervision platform, which mainly includes:

[0007] A communication bus, used to realize the connection and communication between the processor and the memory;

[0008] Memory for storing computer programs;

[0009] A processor is used to execute the computer program to implement the supervision of medical data according to one of the methods of claims 1 to 8. The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0010] The present invention discloses a medical data supervision method, platform and medical server, aiming to solve the problems of complex data elements and difficult hierarchical decision-making in existing medical data management. To address this problem, the present invention first uses a clustering algorithm to group medical data according to similar features, and then extracts key semantic elements of each group of unstructured texts through natural language processing technology and expresses them in a structured manner. Next, association rules are used to mine and analyze the associations between different elements in structured data, and to extract association features that have an important impact on hierarchical decision-making. At the same time, the present invention constructs a feature set that comprehensively covers data importance, sensitivity and application risks, and optimizes it using a feature selection algorithm to obtain a feature subset with significant impact and low correlation. After this subset is converted into a structured vector, it is input into a pre-trained classification model, which is constructed by algorithms such as a support vector machine, and finally an intelligent decision-making model is obtained. The model makes hierarchical decisions on medical data, and generates protective measures for data of different levels in combination with laws, regulations and industry standards. The present invention effectively realizes the refined hierarchical management of medical data and improves data security and application efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 The present invention is a flowchart of a medical data supervision method.

[0012] Figure 2 A schematic diagram of a medical data supervision method of the present invention.

[0013] Figure 3 A schematic diagram of a medical data supervision platform of the present invention.

[0014] Figure 4 It is a structural schematic diagram of a medical data supervision server of the present invention. DETAILED DESCRIPTION

[0015] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0016] like Figure 1-4 In this embodiment, a medical data supervision method, platform and medical server may specifically include:

[0017] Step S101, obtaining a medical data set, and performing clustering processing on the medical data set using a clustering algorithm to obtain multiple data subsets with similar characteristics.

[0018] Obtain a massive amount of medical data sets, which include structured data and unstructured data; perform exploratory analysis on the medical data sets to obtain the distribution characteristics and missing value conditions of the medical data sets; preprocess the medical data sets based on the distribution characteristics and missing value conditions, and the preprocessing includes data cleaning, missing value processing, and outlier processing; based on the characteristics of the medical data sets, use a preset data representation method to convert unstructured data into a structured feature vector; select a preset clustering algorithm to perform clustering analysis on the medical data sets, and divide the medical data sets into at least two data subsets; use a preset evaluation indicator to evaluate the results of the clustering analysis, and adjust the parameters of the clustering algorithm according to the evaluation results until the optimal clustering results are obtained; use a preset association rule mining algorithm to mine medical patterns and laws from the data subsets to obtain the correlation and causality in the medical data sets; apply the clustering results and association rules to the medical field to assist doctors in making clinical decisions.

[0019] Specifically, exploratory data analysis is the first step to understand medical data. For example, a medical record data set of 1,000 patients was collected, which contained information such as age, gender, blood pressure, blood sugar, cholesterol level, and whether they had diabetes. By calculating statistics such as the mean, median, and standard deviation, we can understand the distribution of these indicators. For example, the mean age is 55 years old and the standard deviation is 15 years old, indicating that the age distribution of patients is relatively dispersed; while the mean blood sugar level is 6.5mmol / L and the standard deviation is 1.2mmol / L, indicating that the distribution of blood sugar levels is relatively concentrated. In addition, it was found that there were 5% missing values ​​in the blood pressure data, which may be caused by missing records or equipment failure. This exploratory analysis provides an important basis for subsequent data cleaning and processing. For example, when dealing with missing values, the mean or median can be used to fill in according to the blood pressure values ​​of other patients. The core of data preprocessing is to improve data quality. For the above data set, the missing values ​​are processed first. For example, if the patient has a blood pressure record at other time points, the average blood pressure value of the patient can be used to fill in; if the patient has no other blood pressure records, the median blood pressure of the entire data set can be used to fill in to reduce data deviation. For abnormal values, such as a patient's blood sugar value as high as 20mmol / L, which is far higher than the normal range, it can be regarded as an abnormal value and further checked to see if it is a record error or a special case, and notify the human to handle it accordingly. For data cleaning, for example, if the gender data appears in multiple forms such as "Male", "Female", "Male", "Female", etc., it needs to be uniformly converted into "Male" and "Female" to ensure data consistency. Data representation is to convert unstructured data into a structured form that can be understood by computers. For example, if the medical record contains a text description of the patient's medical history, the TF-IDF method can be used for vectorization. TF-IDF assigns weights to each word based on the frequency of the word in the document and the frequency in the entire corpus. For example, if the word "hypertension" appears many times in the medical record of a patient and appears less frequently in the medical records of other patients, the weight of this word will be relatively high, so it can be used to represent the characteristics of the patient's condition. For categorical data, such as gender, One-Hot encoding can be used to encode "Male" as [1,0] and "Female" as [0,1]. In this way, all data are converted into numerical features, which can be used for subsequent clustering and modeling. Cluster analysis can group similar patients. For example, using the K-means algorithm, the data of the above 1,000 patients are clustered into 3 groups. The K-means algorithm first randomly selects 3 center points, and then assigns each patient to the center point closest to it to form 3 clusters.After that, update the center point of each cluster to the average feature value of all patients in the cluster, and repeat the above steps until the center point no longer changes, or the range of change is less than 5% of the initial transformation range. The clustering results show that the average age of the first group of patients is higher, and their blood sugar and blood pressure are also higher, which may be elderly diabetic patients; the indicators of the second group of patients are relatively normal, and they may belong to the healthy population; the blood sugar and cholesterol levels of the third group of patients are higher, and they may belong to the high-risk population. Through clustering, we have a better understanding of the distribution of patients. The silhouette coefficient can be used to evaluate the clustering effect. The value range of the silhouette coefficient is -1 to 1. The larger the value, the better the clustering effect. For example, if the silhouette coefficient is 0.7, it means that the clustering effect is good; if the silhouette coefficient is 0.2, it means that the clustering effect is poor, and it may be necessary to adjust the clustering parameters, such as the number of clusters K of K-means, or try other clustering algorithms. Association rule mining can discover the correlation between patient characteristics. For example, use the Apriori algorithm to mine association rules in the clustered data set. It was found that in the group of elderly diabetic patients, there is a strong correlation between "hypertension" and "hyperglycemia", that is, the proportion of patients with both hypertension and hyperglycemia is relatively high. This association rule can provide a diagnostic reference for doctors. Finally, the results of clustering and association rules can be used to assist doctors in clinical decision-making. For example, a decision tree algorithm can be used to build a diabetes risk prediction model based on clustering results and association rules. If a patient belongs to the group of elderly diabetic patients and also suffers from hypertension, then the patient's risk of diabetes will be relatively high. According to the prediction results of the diabetes risk prediction model, personalized health management suggestions can be provided to patients.

[0020] Step S102, using natural language processing technology to extract key semantic elements of unstructured text data in the data subset, and achieving structured representation through named entity recognition and part-of-speech tagging methods to obtain a structured data subset.

[0021] Obtain a set of unstructured text data to be processed, and divide the data set into several data subsets; for each data subset, use a word segmentation tool to perform word segmentation on the text data therein to obtain a word segmentation result; based on the word segmentation result, use a part-of-speech tagging function to perform part-of-speech tagging on each word to obtain a part-of-speech tagging result; based on the part-of-speech tagging result, use a named entity recognition function to identify named entities in the text to obtain a named entity recognition result; based on the word segmentation result, the part-of-speech tagging result and the named entity recognition result, use a TF-IDF algorithm to calculate the importance weight of each word, extract several words with the highest weight as key semantic elements, and obtain a set of key semantic elements; based on the set of key semantic elements, construct a structured semantic representation model, use a word vector to represent each key semantic element, and obtain a structured semantic representation through clustering of word vectors; store the structured semantic representation model as structured data, use a relational database for storage, and obtain a structured data subset.

[0022] Specifically, for the unstructured text data in the medical data subset, the jieba word segmentation tool is first used for word segmentation. For example, for a patient medical record "Patient Wang, male, 65 years old, went to the doctor for hypertension and diabetes", after jieba word segmentation, "Patient / Wang / , / male / , / 65 / years old / , / due to / hypertension / and / diabetes / visited the doctor" can be obtained. This step divides the continuous text into meaningful word units, laying the foundation for subsequent processing. Then, each word segmentation is annotated using the part-of-speech tagging function of jieba. In the above example, the following result may be obtained: "Patient / nWang / nr, / wmale / n, / w65 / m years old / q, / wdue to / phypertension / n and / cdiabetes / n went to the doctor / v". Among them, n represents a noun, nr represents a person's name, w represents a punctuation mark, m represents a numeral, q represents a quantifier, p represents a preposition, c represents a conjunction, and v represents a verb. Part-of-speech tagging helps to understand the grammatical function of words in sentences. Then, use Jieba's named entity recognition function to identify named entities in the text. In this example, "Wang" may be identified as a person name entity, and "hypertension" and "diabetes" may be identified as disease entities. Named entity recognition can help extract key information in the text, such as patient name, disease name, etc. Based on the above processing results, the TF-IDF algorithm is used to calculate the importance weight of each word. TF-IDF considers term frequency (TermFrequency) and inverse document frequency (InverseDocumentFrequency), and can effectively identify words that contribute more to the semantics of the text. For example, in a large number of medical record texts, "hypertension" and "diabetes" may obtain higher TF-IDF values, while common words such as "patient" and "visit" may obtain lower values. By setting a threshold or selecting the top-ranked words, a set of key semantic elements can be extracted. Using the extracted key semantic elements, a structured semantic representation model is constructed. Here, word vector technology, such as Word2Vec or GloVe, can be used to map each key semantic element to a high-dimensional vector space. For example, "hypertension" may be represented as a 300-dimensional vector. By calculating the cosine similarity between word vectors, we can find that words with similar semantics, such as "hypertension" and "high blood pressure", may have a high similarity. Finally, the constructed semantic representation model is stored as structured data. You can choose a relational database such as MySQL for storage. For example, you can create a table containing fields such as "word", "part of speech", "entity type", "TF-IDF value", "word vector", etc. This structured storage method facilitates subsequent query and analysis. Through this series of processing, unstructured medical text data is converted into a structured form, which not only retains the semantic information of the original text, but also facilitates computer processing and analysis.This is of great significance for the mining and utilization of medical data, and can support applications such as automatic classification of medical records, retrieval of similar cases, and medical decision support.

[0023] Step S103, using an association rule mining algorithm to analyze the intrinsic connections between different types of data elements in the structured data subset, and obtaining association features that have an important impact on the grading decision.

[0024] Obtain a structured data subset, use data preprocessing technology to clean and transform the structured data subset, and obtain preprocessed transactional data; use association rule mining algorithm to analyze the preprocessed transactional data to obtain frequent item sets and association rules; according to the evaluation indicators of association rules, screen out strong association rules and obtain candidate features; combined with business understanding, select association rules related to the target variable from the candidate features to obtain association features that have an important impact on hierarchical decision-making; use the association features that have an important impact on hierarchical decision-making as input, and use the decision tree algorithm to build a hierarchical decision model; apply the hierarchical decision model to actual business scenarios, predict the new structured data subset, and obtain hierarchical decision results.

[0025] Specifically, obtaining a subset of structured data is the starting point for data analysis. Taking the electronic medical record system of a hospital as an example, structured data containing fields such as basic patient information, diagnosis results, and medication records can be extracted from it. These data usually have quality problems, such as missing values ​​and outliers. These problems need to be dealt with in the data preprocessing stage. For example, the outlier value "999" in the age field may be caused by an entry error, and it can be replaced with the median of the field. For continuous variables such as blood pressure, standardization or normalization methods can be used to convert them so that data of different dimensions are comparable. The preprocessed data can be converted into transactional data, that is, each record represents a visit and contains multiple items (such as symptoms, diagnosis, medication, etc.). Association rule mining algorithms such as Apriori can discover frequent item sets and association rules from these transactional data. For example, it may be found that "hypertension" and "diabetes" often appear at the same time, forming a frequent item set. Further, the rule "If you have high blood pressure, there is a 60% chance that you also have diabetes" may be obtained. The strength of association rules is usually evaluated by indicators such as support, confidence, and lift. Support reflects the frequency of rule occurrence, confidence indicates the reliability of the rule, and lift measures the relevance of the rule. For example, by setting the support threshold to 5% and the confidence threshold to 80%, strong association rules can be screened out. These rules constitute the candidate feature set. Combined with the business understanding of medical experts, rules related to the target variable (such as hospitalization risk) can be selected from the candidate features. For example, the rule "hypertension + diabetes → increased risk of hospitalization" may be selected as an important association feature. This feature selection process not only relies on statistical indicators, but also needs to consider medical knowledge and clinical experience. Using the screened association features as input, a decision tree model can be constructed to make hierarchical decisions. The advantage of decision trees is that they are highly interpretable and suitable for decision support in medical scenarios. For example, the model may generate a tree in which the first branch point is "whether you have both hypertension and diabetes", which leads to different risk levels based on the answer. Finally, the constructed decision tree model is applied to new patient data. Suppose there is a new patient. The system will traverse each node of the decision tree based on the patient's examination results and past medical history, and finally derive the patient's hospitalization risk level. This hierarchical decision result can assist doctors in making clinical decisions, such as determining the frequency of follow-up or formulating personalized treatment plans. This series of steps constitutes a complete medical data mining and decision support process. From the original structured data to the final decision model, each step is extracting valuable information. This not only improves the utilization efficiency of medical resources, but also provides data support for precision medicine. However, it should be noted that the results of the model should be used as an auxiliary tool, and the final medical decision still needs to be combined with the doctor's professional judgment.

[0026] Step S104, based on the preset medical data grading evaluation index system, construct a feature set that comprehensively covers various data elements in the medical data set from the dimensions of data importance, sensitivity and application risk.

[0027] Obtain a medical data set, and extract the characteristics of various data elements in the medical data set from the three dimensions of data importance, data sensitivity and application risk according to the preset medical data grading evaluation index system; adopt a feature selection algorithm to screen out feature subsets that have a great impact on the grading evaluation results according to the correlation between the features and the grading evaluation indicators; for each feature in the feature subset, according to its data type and value range, adopt a data discretization method to convert the continuous feature into a discrete feature; according to the business meaning of each feature in the feature subset, divide the features into three dimensions of data importance, data sensitivity and application risk, and construct a feature set that comprehensively covers the medical data set; adopt a classification algorithm, take the feature set as input, and take the medical data grading evaluation index as output, and train the medical data grading evaluation model; apply the medical data grading evaluation model to the new medical data set, and perform grading evaluation on various data elements therein to obtain the importance, sensitivity and application risk level of each data element; according to the grading evaluation results of the data elements, formulate corresponding data management strategies and application specifications to realize the grading management and compliant application of medical data.

[0028] Specifically, first, obtain a medical data set, such as the electronic medical record system of a tertiary hospital. The preset evaluation index system includes data importance (such as the impact of diagnostic results on treatment plans), data sensitivity (such as the sensitivity of patient privacy information) and application risk (such as the possible consequences of data leakage). Extract features from these dimensions. For example, for diagnostic results, features such as disease severity and treatment difficulty can be extracted; for patient information, features such as age and occupation can be extracted; for medication records, features such as drug category and dosage can be extracted. Use feature selection algorithms such as recursive feature elimination to filter out feature subsets with greater influence based on the correlation between features and evaluation indicators. For the filtered feature subsets, use data discretization methods. For example, convert the continuous feature of age into a discrete feature: 0-18 years old is a minor, 19-65 years old is an adult, and over 66 years old is an elderly. This conversion helps improve the generalization ability and computational efficiency of the model. According to the business meaning of the feature, it is divided into three dimensions. For example, the accuracy of the diagnostic result belongs to the data importance dimension, the patient's ID number belongs to the data sensitivity dimension, and the risk of drug side effects belongs to the application risk dimension. This division helps to comprehensively evaluate all aspects of medical data. Classification algorithms such as random forests are used to train medical data hierarchical evaluation models with feature sets as input and evaluation indicators as output. The advantage of random forests is that they can handle high-dimensional features and are not prone to overfitting. During model training, cross-validation can be used to evaluate model performance to ensure the stability and generalization ability of the model. The trained model is applied to new medical data sets to perform hierarchical evaluations on various data elements. For example, the model may rate the patient's pathological examination results as high importance, medium sensitivity, and low application risk. This multi-dimensional evaluation result provides a basis for formulating data management strategies. According to the evaluation results, corresponding data management strategies and application specifications are formulated. For example, for high-importance data, multiple backup strategies can be adopted; for high-sensitivity data, strict access control and encryption measures can be implemented; for high-application risk data, detailed use review processes can be formulated. The implementation of these strategies and specifications helps to maximize the value of medical data while protecting patient privacy. Through this series of steps, medical institutions can achieve refined management of data, improve data utilization efficiency, and reduce the risk of data leakage and abuse. This will not only help improve the quality of medical services, but also provide reliable data support for medical research and public health decision-making.

[0029] Step S105, using a feature selection algorithm to perform feature selection on the feature set, taking the classification index as the optimization target, balancing the comprehensiveness and pertinence of the feature set, and obtaining a feature subset with significant impact and low correlation.

[0030] Obtain a preset medical data grading evaluation index system and a constructed feature set, and use the grading index as the optimization target for feature selection; use a feature selection algorithm based on mutual information to calculate the mutual information value between each feature and the grading index, and the larger the mutual information value, the more significant the impact of the feature on the grading result; calculate the average correlation within the feature subset, select feature combinations with low correlation, and reduce the redundancy between features; balance the comprehensiveness and pertinence of the feature subset by setting the mutual information threshold and the correlation threshold, and obtain a feature subset that has both significant impact and low correlation; according to the data type and value distribution of each feature in the feature subset, use the data normalization method to scale and standardize the feature data; input the processed feature subset into the classification algorithm to train the medical data grading evaluation model; use the cross-validation method to evaluate the model, adjust the feature selection threshold according to the evaluation results, and obtain the feature subset and grading model with the best performance.

[0031] Specifically, the medical data grading evaluation index system is a key tool for evaluating the importance, sensitivity and application risk of medical data. For example, basic information of patients such as age and gender may be assigned lower sensitivity, while HIV test results have extremely high sensitivity. This preset index system provides a clear optimization goal for feature selection. The feature selection algorithm based on mutual information evaluates the importance of features by calculating the mutual information value between features and grading indicators. Assuming there is a data set containing features such as patient age, blood pressure, and blood sugar, the mutual information calculation may show that the mutual information value between blood pressure and cardiovascular disease risk grading is the highest, indicating that blood pressure is an important indicator for predicting cardiovascular disease risk. The average correlation calculation within the feature subset helps to reduce redundancy. For example, weight and BMI index may be highly correlated, and selecting one of them as a representative can simplify the model and improve efficiency. By setting the mutual information threshold (such as 0.5) and the correlation threshold (such as 0.7), feature combinations that have a significant impact on the grading results and are independent of each other can be screened out. Data normalization is essential for features of different scales. For example, converting age (0-100) and blood pressure (90-180 mmHg) to the same range of 0-1 ensures that they have the same weight in the model. This processing can improve the stability and accuracy of the model. In terms of the choice of classification algorithm, decision tree models are popular in the medical field because of their interpretability. For example, a decision tree for predicting diabetes risk may first consider fasting blood glucose levels, then age and BMI index, forming an intuitive decision path. Support vector machines may perform better when processing high-dimensional data, especially when predicting complex diseases such as cancer. Cross-validation is an effective way to evaluate model performance. Through 10-fold cross-validation, the performance of the model on different data subsets can be obtained. If it is found that the model performs poorly on certain subsets, it may be necessary to re-examine the feature selection process and adjust the mutual information or correlation threshold. For example, if the accuracy of the model in the elderly patient group is significantly lower than the overall level, it may be necessary to add some features specifically for the elderly. Through this iterative optimization process, a model that can accurately predict the classification of medical data and has good generalization ability can be obtained. This model can not only assist doctors in risk assessment, but also provide data support for the allocation of medical resources and the formulation of personalized treatment plans, thereby improving the overall quality of medical services.

[0032] Step S106, converting the feature subset into a structured feature vector and inputting it into a preset classification model. The classification model is obtained by training the support vector machine algorithm through labeled samples to obtain an intelligent decision-making model for hierarchical management of medical data.

[0033] The original medical data is obtained from the medical information system; data cleaning is performed on the original medical data, including missing value filling and outlier removal, to obtain preprocessed data; relevant features are extracted from the preprocessed data according to medical field knowledge to form a feature set; for the numerical features in the feature set, Min-MaxScaler is used to standardize the data, and the numerical values ​​are scaled to the interval [0, 1] to form a standardized feature set; the features in the standardized feature set are arranged in a preset order and converted into a structured feature vector, and the structured feature vector contains numerical and categorical data; a classification model is pre-established, a logistic regression classifier is selected, and model hyperparameters are determined; according to the classification model, a training set and a validation set are divided from the labeled samples, the logistic regression classifier is trained using the training set, and the model hyperparameters are adjusted using the validation set to obtain a trained classifier; the structured feature vector is input into the trained classifier, and the medical data classification result is output, and the medical data classification result is a preset category label; according to the medical data classification result, according to the pre-set management rules, the medical data management operation is determined to realize the medical data classification management.

[0034] Specifically, raw medical data may include basic information of patients, diagnosis records, treatment plans, and examination results. Data cleaning is a key step to ensure data quality. In medical data, missing values ​​may be caused by the patient not having a certain examination or the doctor not recording certain information. For example, for blood pressure data, the historical average of the patient or the average of the same age group may be used to fill it. Outliers may be caused by measurement errors or recording errors, such as body temperature recorded as 50°C. These obviously unreasonable data need to be removed or corrected. Feature extraction is the process of converting raw data into features that can be used for analysis. In the medical field, this may include extracting keywords from medical record text or extracting statistical features from continuous physiological signals. For example, features such as heart rate and heart rate variability can be extracted from electrocardiogram data. Data normalization is an important step to ensure that features of different scales can be compared fairly. Min-MaxScaler scales all numerical features to between 0 and 1. For example, the age range may be 0 to 100, while the blood sugar level may be 70 to 200 mg / dL. After normalization, both features are mapped to the range of 0 to 1, so that they have the same weight in subsequent analysis. The construction of structured feature vectors is to make the data suitable for processing by machine learning algorithms. Numerical data such as age and blood pressure can be used directly, while categorical data such as gender and blood type need to be encoded. For example, gender can be encoded as 0 (female) and 1 (male). Logistic regression is a commonly used classification algorithm, which is particularly suitable for binary classification problems. In medical data classification, it can be used to predict whether a patient belongs to the high-risk group. The hyperparameters of the model, such as regularization strength, can be determined by cross-validation. For example, different regularization strengths such as 0.01, 0.1, 1, and 10 may be tried, and the parameters that perform best on the validation set are selected. The division of the training set and the validation set is usually done by random sampling. For example, 80% of the data may be used as the training set and 20% as the validation set. During the training process, the model learns how to predict the classification results based on the input features. The validation set is used to evaluate the model performance and adjust the hyperparameters to prevent overfitting. The classification results may be binary (such as high risk / low risk) or multi-category (such as low risk / medium risk / high risk). These results directly affect the subsequent data management strategy. For example, patient data classified as high-risk may require more frequent updates and stricter access control. Finally, formulate corresponding data management operations based on the classification results. This may include data storage policies (such as high-risk data is stored on more secure servers), access control (such as only doctors of a certain level can view the complete records of high-risk patients), data update frequency (such as high-risk patient data is updated daily, low-risk patient data is updated weekly), etc. This hierarchical management approach can optimize the allocation of medical resources and improve data management efficiency while ensuring patient privacy and data security.

[0035] Step S107, using an intelligent decision-making model to make hierarchical decisions on medical data sets, and formulating protection measures for different levels of data in combination with laws, regulations and industry standards on medical data security and privacy protection.

[0036] The original medical data is obtained, and a support vector machine model is used to preliminarily classify the original medical data according to the pre-established correspondence between the type of medical data and the degree of sensitivity to obtain a first-level data set; it is determined whether the first-level data set contains sensitive privacy information such as the patient's name, ID number and contact information. If so, a secondary classification is performed according to the degree of sensitivity to obtain a second-level data set; if not, the first-level data set is classified as hierarchical data; for the first-level data set and the second-level data set, the corresponding security level is determined according to relevant laws and regulations or confidentiality agreements, and the security level includes three levels: high, medium and low; if the security level is high, the AES encryption algorithm is used to encrypt and store the data, and a role-based access control strategy is implemented to obtain an encrypted data set; if the security level is medium, a generalized algorithm is used to desensitize the data, and an audit log is recorded to obtain a desensitized data set; if the security level is low, permission management is implemented and access records are recorded.

[0037] Specifically, first, the original medical data is obtained from the hospital information system, which may include the patient's basic information, diagnosis records, treatment plans, etc. The support vector machine model is used for preliminary classification. For example, a patient's blood pressure, blood sugar, and age characteristics may be used as input, and the model classifies them as "general health information" based on the pre-established correspondence. Next, the system checks whether there is sensitive privacy information. Assuming that the patient's ID number is found in the first-level data set, the system will classify it as highly sensitive information for secondary classification. This hierarchical approach can manage data of different sensitivities more finely. According to relevant regulations or confidentiality agreements, the system assigns security levels to data. High-security data, such as patients' HIV test results, are protected using the AES encryption algorithm. AES is a symmetric encryption algorithm that uses the same key for encryption and decryption, which can effectively prevent unauthorized access. At the same time, role-based access control is implemented, such as only attending doctors and specific nurses can access these sensitive information. For medium-security data, such as the patient's general medical history, the system uses a generalization algorithm for desensitization. For example, replace specific ages with age groups (such as "30-40 years old"), or summarize precise residential addresses as regions (such as "city center"). This approach protects personal privacy while still retaining the analytical value of the data. Audit log records ensure the traceability of data access and help discover potential security vulnerabilities. Low-security data, such as general statistical information of hospitals, are mainly protected through permission management and access records. This may include setting different levels of user permissions, such as interns can only view but not modify certain data. The maintenance of access records helps monitor data usage and promptly detect abnormal access patterns. This multi-level data classification management approach not only meets legal requirements, but also strikes a balance between protecting patient privacy and promoting medical research. For example, de-identified intermediate-level security data can be used for epidemiological research without leaking personal identity information. At the same time, highly sensitive data is protected with the strictest protection, ensuring the trust of patients and the reputation of medical institutions. In this way, medical institutions can manage and utilize valuable medical data resources more effectively, while also providing patients with a safer and more reliable medical service environment. 。

[0038] The present invention provides a medical data supervision platform, which mainly includes: a data acquisition module, which is used to acquire a medical data set, and cluster the medical data set using a clustering algorithm to obtain a plurality of data subsets with similar characteristics; a clustering processing module, which is used to extract key semantic elements of unstructured text data in the data subset using natural language processing technology, and achieve structured representation through named entity recognition and part-of-speech tagging methods to obtain structured data subsets; a semantic extraction module, which is used to analyze the intrinsic connections between different types of data elements in the structured data subset using an association rule mining algorithm to obtain association features that have an important impact on hierarchical decision-making; an association analysis module, which is used to analyze the importance, sensitivity, and other aspects of the data according to a preset medical data hierarchical evaluation index system. The module is used to construct a feature set that comprehensively covers all types of data elements in the medical data set based on the degree and application risk dimensions; the feature construction module is used to select features from the feature set using a feature selection algorithm, taking the grading index as the optimization goal, balancing the comprehensiveness and pertinence of the feature set, and obtaining a feature subset with significant impact and low correlation; the feature selection module is used to convert the feature subset into a structured feature vector, which is input into a preset classification model. The classification model is obtained by training the support vector machine algorithm through labeled samples to obtain an intelligent decision-making model for hierarchical management of medical data; the decision generation module is used to make hierarchical decisions on the medical data set using an intelligent decision-making model, and formulate protection measures for data of different levels in combination with the laws, regulations and industry standards on medical data security and privacy protection.

[0039] The present invention provides a medical data supervision server, comprising:

[0040] A communication bus, used to realize the connection and communication between the processor and the memory;

[0041] Memory for storing computer programs;

[0042] A processor for executing the computer program to implement the supervision of medical data according to the method of any one of claims 1-8.

[0043] It should be noted that the above examples are only some specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and there are many variations. All variations that can be directly derived or associated with the content disclosed by a person skilled in the art should be considered as the protection scope of the present invention.

Claims

1. A medical data supervision method, characterized in that: include: Acquire a medical data set, and use a clustering algorithm to perform clustering processing on the medical data set to obtain multiple data subsets with similar characteristics; Natural language processing technology is used to extract key semantic elements of unstructured text data in the data subset, and structured representation is achieved through named entity recognition and part-of-speech tagging methods to obtain a structured data subset; association rule mining algorithms are used to analyze the inherent connections between different types of data elements in the structured data subset to obtain association features that have an important impact on hierarchical decision-making; Based on the preset medical data grading evaluation indicator system, a feature set that comprehensively covers all types of data elements in the medical data set is constructed from the dimensions of data importance, sensitivity and application risk; A feature selection algorithm is used to select features from feature sets, with grading indicators as optimization targets, to balance the comprehensiveness and pertinence of feature sets, and obtain feature subsets with significant impact and low correlation; the feature subsets are converted into structured feature vectors and input into a preset classification model. The classification model is obtained by training labeled samples using a support vector machine algorithm to obtain an intelligent decision-making model for hierarchical management of medical data; an intelligent decision-making model is used to make hierarchical decisions on medical data sets, and protective measures are formulated for data of different levels in combination with laws, regulations and industry standards for medical data security and privacy protection.

2. The method according to claim 1, characterized in that The medical data set is obtained, and a clustering algorithm is used to perform clustering processing on the medical data set to obtain multiple data subsets with similar characteristics, including: Obtain a massive amount of medical data collection, including structured data and unstructured data; Conduct exploratory analysis on medical data sets to obtain the distribution characteristics and missing value conditions of medical data sets; According to the distribution characteristics and missing values, the medical data set is preprocessed, including data cleaning, missing value processing and outlier processing; According to the characteristics of medical data sets, a preset data representation method is used to convert unstructured data into structured feature vectors; Selecting a preset clustering algorithm, performing cluster analysis on the medical data set, and dividing the medical data set into at least two data subsets; Use the preset evaluation indicators to evaluate the results of clustering analysis, and adjust the parameters of the clustering algorithm according to the evaluation results until the optimal clustering result is obtained; Using the preset association rule mining algorithm, we can mine medical patterns and rules from the data subsets to obtain the correlation and causality in the medical data set; The clustering results and association rules are applied in the medical field to assist doctors in making clinical decisions.

3. The method according to claim 1, characterized in that The method uses natural language processing technology to extract key semantic elements of unstructured text data in the data subset, and implements structured representation through named entity recognition and part-of-speech tagging methods to obtain a structured data subset, including: Obtaining an unstructured text data set to be processed, and dividing the data set into a number of data subsets; For each data subset, use the word segmentation tool to segment the text data therein to obtain the word segmentation results; According to the word segmentation results, the part-of-speech tagging function is used to tag each word to obtain the part-of-speech tagging results; According to the part-of-speech tagging results, the named entity recognition function is used to identify the named entities in the text and obtain the named entity recognition results; According to the word segmentation results, part-of-speech tagging results and named entity recognition results, the TF-IDF algorithm is used to calculate the importance weight of each word, and several words with the highest weight are extracted as key semantic elements to obtain a set of key semantic elements; According to the set of key semantic elements, a structured semantic representation model is constructed, and word vectors are used to represent each key semantic element. The structured semantic representation is obtained by clustering word vectors. The structured semantic representation model is stored as structured data in a relational database to obtain a structured data subset.

4. The method according to claim 1, characterized in that The association rule mining algorithm is used to analyze the intrinsic relationship between different types of data elements in the structured data subset to obtain association features that have an important impact on the grading decision, including: Obtain a structured data subset, and use data preprocessing technology to clean and transform the structured data subset to obtain preprocessed transactional data; The association rule mining algorithm is used to analyze the pre-processed transactional data to obtain frequent item sets and association rules; According to the evaluation index of association rules, strong association rules are screened out to obtain candidate features; Combined with business understanding, association rules related to the target variable are selected from candidate features to obtain association features that have an important impact on grading decisions; The associated features that have an important impact on the hierarchical decision are used as input, and the hierarchical decision model is constructed using the decision tree algorithm; The hierarchical decision model is applied to actual business scenarios, and predictions are made on new structured data subsets to obtain hierarchical decision results.

5. The method according to claim 1, characterized in that According to the preset medical data grading evaluation index system, a feature set that comprehensively covers various data elements in the medical data set is constructed from the dimensions of data importance, sensitivity and application risk, including: Obtain a medical data set, and extract the characteristics of various data elements in the medical data set from three dimensions: data importance, data sensitivity, and application risk, based on a preset medical data grading evaluation indicator system; Using feature selection algorithm, according to the correlation between features and grading evaluation indicators, the feature subset with great influence on grading evaluation results is screened out; For each feature in the feature subset, according to its data type and value range, the data discretization method is used to convert the continuous feature into a discrete feature; According to the business meaning of each feature in the feature subset, the features are divided into three dimensions: data importance, data sensitivity, and application risk, and a feature set that fully covers the medical data set is constructed; A classification algorithm is used to train a medical data grading evaluation model with a feature set as input and a medical data grading evaluation index as output; Apply the medical data hierarchical assessment model to new medical data sets, conduct hierarchical assessments on various data elements, and obtain the importance, sensitivity, and application risk level of each data element; Based on the hierarchical assessment results of data elements, corresponding data management strategies and application specifications are formulated to achieve hierarchical management and compliant application of medical data.

6. The method according to claim 1, characterized in that The feature selection algorithm is used to select features from the feature set, with the classification index as the optimization target, to balance the comprehensiveness and pertinence of the feature set, and obtain a feature subset with significant influence and low correlation, including: Obtain the preset medical data grading evaluation index system and the constructed feature set, and use the grading index as the optimization target of feature selection; A feature selection algorithm based on mutual information is used to calculate the mutual information value between each feature and the grading index. The larger the mutual information value, the more significant the impact of the feature on the grading result. Calculate the average correlation within the feature subset, select the feature combination with low correlation, and reduce the redundancy between features; By setting the mutual information threshold and correlation threshold, the comprehensiveness and pertinence of the feature subset are balanced, and a feature subset with significant impact and low correlation is obtained; According to the data type and value distribution of each feature in the feature subset, the feature data is scaled and standardized using the data normalization method; Input the processed feature subset into the classification algorithm to train the medical data grading assessment model; The model was evaluated using the cross-validation method, and the feature selection threshold was adjusted based on the evaluation results to obtain the feature subset and classification model with the best performance.

7. The method according to claim 1, characterized in that The feature subset is converted into a structured feature vector and input into a preset classification model. The classification model is obtained by training the support vector machine algorithm through labeled samples to obtain an intelligent decision-making model for hierarchical management of medical data, including: Obtaining raw medical data from medical information systems; For the original medical data, data cleaning is performed, including missing value filling and outlier removal, to obtain preprocessed data; From the preprocessed data, relevant features are extracted based on medical field knowledge to form a feature set; For the numerical features in the feature set, Min-MaxScaler is used to standardize the data and scale the values ​​to the interval [0, 1] to form a standardized feature set; Arrange the features in the standardized feature set in a preset order and convert them into a structured feature vector, which contains numerical and categorical data; Pre-build a classification model, select a logistic regression classifier, and determine the model hyperparameters; According to the classification model, the labeled samples are divided into training set and validation set, the training set is used to train the logistic regression classifier, and the validation set is used to adjust the model hyperparameters to obtain the trained classifier; Input the structured feature vector into the trained classifier and output the medical data classification result, where the medical data classification result is a preset category label; According to the medical data classification results and in accordance with pre-set management rules, medical data management operations are determined to achieve medical data classification management.

8. The method according to claim 1, characterized in that The intelligent decision-making model is used to make hierarchical decisions on medical data sets, and combined with the laws, regulations and industry standards for medical data security and privacy protection, protection measures for different levels of data are formulated, including: Obtaining original medical data, and using a support vector machine model to preliminarily classify the original medical data according to the pre-established correspondence between medical data types and sensitivity levels, to obtain a first-level data set; Determine whether the first-level data set contains sensitive privacy information such as the patient's name, ID number, and contact information. If so, perform secondary classification based on the sensitivity to obtain the second-level data set; If not included, the first-level data are combined as hierarchical data; For the first-level data set and the second-level data set, the corresponding security level is determined according to relevant laws and regulations or confidentiality agreements. The security level includes high, medium and low levels; If the security level is high, the AES encryption algorithm is used to encrypt and store the data, and a role-based access control strategy is implemented to obtain an encrypted data set; If the security level is medium, a generalized algorithm is used to desensitize the data, and the audit log is recorded to obtain a desensitized data set; If the security level is low, permission management is implemented and access records are recorded.

9. A medical data supervision platform, characterized in that: The platform includes: A data acquisition module is used to acquire a medical data set, and cluster the medical data set using a clustering algorithm to obtain multiple data subsets with similar characteristics; The clustering processing module is used to extract the key semantic elements of the unstructured text data in the data subset by using natural language processing technology, and to achieve structured representation through named entity recognition and part-of-speech tagging methods to obtain a structured data subset; the semantic extraction module is used to analyze the intrinsic connection between different types of data elements in the structured data subset by using association rule mining algorithms to obtain association features that have an important impact on hierarchical decision-making; The association analysis module is used to construct a feature set that comprehensively covers various data elements in the medical data set from multiple dimensions such as data importance, sensitivity and application risk according to the preset medical data grading evaluation index system; the feature construction module is used to use the feature selection algorithm to select features from the feature set, take the grading index as the optimization goal, balance the comprehensiveness and pertinence of the feature set, and obtain a feature subset with significant impact and low correlation; the feature selection module is used to convert the feature subset into a structured feature vector, and input it into the preset classification model. The classification model is obtained by training the support vector machine algorithm through labeled samples to obtain an intelligent decision-making model for the grading management of medical data; the decision generation module is used to use the intelligent decision-making model to make grading decisions on the medical data set, and formulate protection measures for data of different levels in combination with the laws, regulations and industry standards for medical data security and privacy protection.

10. A medical data supervision server, characterized in that: The server comprises: A communication bus, used to realize the connection and communication between the processor and the memory; Memory for storing computer programs; A processor for executing the computer program to implement the supervision of medical data according to the method of any one of claims 1-8.

Images