AI-based heart disease prediction system

Abstract

The present invention relates to the field of cardiac data analysis technology, specifically to an AI-based cardiac disease prediction system, the system comprising: a feature relationship modeling module maps each type of information into a multidimensional feature space based on the collected cardiac data, analyzes the degree of conditional correlation between feature nodes, connects the feature nodes to each other through correlation information, quantifies the linkage effect between features, and obtains feature conditional dependency analysis results. In the present invention, various types of cardiac health data are noded by constructing a multidimensional feature space, and an association structure is established based on the conditional correlation between data features. The synergy and mutual dependence between features are revealed. By analyzing the joint state probability of these feature nodes, the linkage effect of feature combinations on the overall risk is quantified, thereby enhancing the depth of analysis of disease risks. In addition, by comparing the independent state probability of a single feature with the joint state probability of a feature combination, feature combinations that contribute significantly to the risk can be identified.

Description

Technical Field

[0001] The present invention relates to the technical field of cardiac data analysis, and in particular to an AI-based cardiac disease prediction system. Background Art

[0002] Cardiac data analysis is to identify heart health conditions, disease risk factors and early symptoms through the mining, analysis and processing of heart health-related data. The field covers a variety of data analysis technologies, including machine learning, data mining, bioinformatics, etc., and provides accurate health assessments by analyzing multi-source data from electrocardiograms, echocardiograms, blood tests and daily activities.

[0003] Among them, the heart disease prediction system uses heart data analysis technology to predict an individual's future risk of heart disease. The system analyzes the user's physical data, such as age, blood pressure, cholesterol level, heart rate and other indicators, and combines machine learning models and other algorithms to provide disease risk scores and predictions to help users detect potential heart health problems early. It provides users with early warnings of heart disease, supports users to make lifestyle adjustments or medical examinations early, reduces the risk of heart attacks, and provides data support for doctors in the decision-making process.

[0004] In the prediction of heart disease risk, existing technologies have difficulty in conducting in-depth analysis of the complex relationships between features, which makes it difficult for the model to identify potential synergies and interdependencies between features. For example, it is difficult to determine the combined effect of high cholesterol and high blood pressure on the increased risk of heart disease, which will result in an incomplete and inaccurate assessment of patient risk and may underestimate the aggravating effect of combined features on disease risk, thereby affecting the effectiveness and accuracy of early warning. The analysis model of existing technologies fails to effectively quantify the contribution of feature nodes and the linkage effect between adjacent nodes, which may lead to prediction deviations and is not comprehensive when evaluating the overall impact of feature combinations. For example, it is difficult to accurately quantify the specific contribution of the combination of blood pressure and heart rate to risk, which causes the prediction results to deviate from the actual situation and miss high-risk individuals. In addition, existing technologies are difficult to reflect potential changes in the disease, which is not conducive to patients adjusting their lifestyles or receiving medical examinations in a timely manner when the risk of disease changes. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose an AI-based heart disease prediction system.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: The AI-based heart disease prediction system includes:

[0007] The feature relationship modeling module maps each type of information into a multidimensional feature space based on the collected cardiac data, analyzes the degree of conditional correlation between feature nodes, connects feature nodes to each other through correlation information, quantifies the linkage effect between features, and obtains the feature conditional dependency analysis results;

[0008] The spatial feature distribution analysis module performs a detailed analysis on the distribution of feature nodes in the multidimensional feature space based on the feature condition dependency analysis results, determines the synergy of feature combinations, quantifies the contribution of adjacent nodes and connecting edges in the feature combinations, divides them into influential feature combinations, and constructs a feature cluster dependency structure;

[0009] The health risk prediction module divides the heart data into fixed windows according to time based on the data of the feature cluster dependency structure, analyzes the state changes of the feature clusters in each time window, quantifies the fluctuation of the feature clusters, obtains the fluctuation index of the feature clusters, judges the stability of the health status according to the fluctuation index, and generates the health risk assessment result;

[0010] Based on the health risk assessment results, the risk group identification module performs cluster analysis on individual data points determined to be at a high risk level, identifies feature clusters with risks, and generates risk group identification results in heart disease data.

[0011] As a further solution of the present invention, the step of obtaining the conditional correlation degree between the analysis feature nodes is specifically as follows:

[0012] Based on the collected heart data, the characteristics of heart disease including age, blood pressure, cholesterol, and heart rate information are extracted and preprocessed, and the processed data are mapped into the multidimensional feature space as feature nodes. Each node contains its feature value and corresponding sample index, and the feature node data set is obtained;

[0013] According to the characteristic node data set, the formula is adopted:

[0014] ;

[0015] Calculated at feature nodes In the event of a The conditional probability , get the analysis results of the correlation between adjacent feature nodes;

[0016] in, is the adjacent feature node and The joint probability of simultaneous occurrence, It is a feature node The marginal probability of The frequency of occurrence in the sample.

[0017] As a further solution of the present invention, the step of obtaining the linkage effect between the quantitative features is specifically:

[0018] According to the analysis result of the correlation degree of the adjacent feature nodes, the connection weight between the nodes is determined. If the weight exceeds the preset correlation threshold, a connection edge is established between the nodes, and the weight value is attached to the edge and recorded in the adjacency matrix or adjacency table to generate a feature association graph;

[0019] According to the node and edge information in the feature association graph, the formula is adopted:

[0020] ;

[0021] Compute feature nodes , , The joint probability of being in certain states at the same time , get the feature condition dependency analysis results;

[0022] in, is the conditional probability, indicating that at the node and Under the conditions that have occurred, the node The probability of occurrence, is the conditional probability, indicating that at the node Under the conditions that have occurred, the node The probability of occurrence, is the edge probability, indicating that the node The probability of appearing in the sample data.

[0023] As a further solution of the present invention, the step of obtaining quantified contributions of adjacent nodes and connecting edges in the feature combination is specifically as follows:

[0024] Based on the feature condition dependency analysis result, by comparing the edge probability of a single feature node with the probability in a joint state, identifying and marking feature combinations with synergistic effects, and obtaining feature combination synergistic effect analysis results;

[0025] According to the analysis results of the synergistic effect of the feature combination, the formula is adopted:

[0026] ;

[0027] Calculating the synergy score , get the overall synergistic effect data of the feature combination;

[0028] in, It is a feature node The weight of is the conditional probability, indicating the feature node On its adjacent nodes The probability of occurrence under given conditions, It is a representation and feature node The set of adjacent feature nodes, It is a feature node The probability of appearing independently in the data set, Used to accumulate the probability of independent occurrence of each feature node , is the total number of feature nodes involved in the feature combination.

[0029] As a further solution of the present invention, the step of acquiring the feature cluster dependency structure is specifically as follows:

[0030] According to the overall synergistic effect data of the feature combination, the synergistic effect score is compared with the preset scoring standard, the feature combination with a score higher than the standard is screened, the corresponding node and edge information is extracted, and the influential feature combination is obtained;

[0031] The information of nodes and edges in the influential feature combination is recorded in an adjacency matrix or an adjacency table according to the weights, nodes and edges are added in sequence, the weight value of each edge is marked, the associated nodes are grouped and merged, and a feature cluster dependency structure is established.

[0032] As a further solution of the present invention, the step of obtaining the fluctuation index of the characteristic cluster is specifically as follows:

[0033] Based on the data of the feature cluster dependency structure, the cardiac data is divided into fixed time windows by time, the node state information in each feature cluster is extracted in each time window, the state changes in adjacent time windows are compared, the fluctuation of the feature cluster is marked, and the state change information of the feature cluster in the time window is obtained;

[0034] According to the feature cluster state change information within the time window, the formula is used:

[0035] ;

[0036] Calculate fuzzy entropy value , get the volatility index of the characteristic cluster;

[0037] in, is the index variable of the feature node, is within the time window The similarity probability of feature nodes, is the number of main feature nodes in the feature cluster, is the weight adjustment coefficient, is the index variable of the feature node, It is The dissimilarity probability of feature nodes, is the number of non-similar feature nodes in the feature cluster.

[0038] As a further solution of the present invention, the steps for obtaining the health risk assessment results are specifically as follows:

[0039] According to the fluctuation index of the characteristic cluster, by comparing the fluctuation index in each time window with the stability threshold, the fluctuation amplitude of the characteristic cluster is determined, and the change trend of the fluctuation index in the time window is monitored to obtain the stability analysis result of the health state;

[0040] According to the stability analysis result of the health status, the fluctuation index in each time window is divided into multiple risk levels, the risk level information of each time window is recorded, and the health risk assessment result is generated.

[0041] As a further solution of the present invention, the steps of obtaining the risk group identification result in the heart disease data are specifically as follows:

[0042] Based on the health risk assessment results, extract individual data points determined to be high risk, identify high-risk feature clusters, organize the value range, fluctuation and abnormal points of each feature cluster into a data set, and screen them to obtain risk feature cluster information;

[0043] According to the risk feature cluster information, the formula is used:

[0044] ;

[0045] Calculate neighborhood density , identifying the areas with the highest density and generating risk group identification results in heart disease data;

[0046] in, represents the number of the feature cluster, It is The number of high-risk data points in a feature cluster, It is The neighborhood volume of feature clusters, is the number of high-risk feature clusters in the neighborhood.

[0047] Compared with the prior art, the advantages and positive effects of the present invention are:

[0048] In the present invention, by constructing a multidimensional feature space, various types of heart health data are noded, and an association structure is established according to the conditional correlation between data features. The synergy and interdependence between features are revealed. By analyzing the joint state probability of these feature nodes, the linkage effect of feature combinations on the overall risk is quantified, and the depth of analysis of disease risk is enhanced. In addition, by refining the distribution of feature nodes in multidimensional space, comparing the independent state probability of a single feature with the joint state probability of a feature combination, it is possible to identify feature combinations that contribute significantly to risk, reveal the strength of the synergistic effect between features, and extract important feature cluster combinations and their association structures. On this basis, the fluctuations of each feature cluster in the time series are dynamically monitored, and the stability of the health status is judged in real time to ensure efficient and accurate tracking of individual risk changes and avoid omissions of high-risk states. The feature cluster density of high-risk individuals is recursively analyzed to aggregate potential high-risk populations, and group characteristics can be quickly identified and a risk description set can be formed. The early warning support for high-risk groups is strengthened, so that individuals can receive timely monitoring suggestions and early warning analysis in the early stage, and provide accurate data support for decision-making assistance. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a system flow chart of the present invention;

[0050] Figure 2 A flow chart of analyzing the degree of conditional correlation between feature nodes in the present invention;

[0051] Figure 3 It is a flow chart of the linkage effect between quantitative features of the present invention;

[0052] Figure 4 A flow chart of quantifying the contribution of adjacent nodes and connecting edges in a feature combination according to the present invention;

[0053] Figure 5 A flow chart for constructing a feature cluster dependency structure for the present invention;

[0054] Figure 6 A flow chart of obtaining the volatility index of the characteristic cluster according to the present invention;

[0055] Figure 7 A flow chart for obtaining health risk assessment results for the present invention;

[0056] Figure 8 A flow chart of obtaining risk group identification results in heart disease data according to the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0058] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0059] See also Figure 1 , the AI-based heart disease prediction system includes:

[0060] The feature relationship modeling module extracts and preprocesses the features of heart disease, including age, blood pressure, cholesterol, and heart rate, based on the collected heart data. Each type of information is mapped to a multidimensional feature space, and the preprocessed information is defined as a feature node in the space. The dependency between adjacent feature nodes is calculated, and the degree of conditional correlation between feature nodes is analyzed. The feature nodes are connected to each other through correlation information to form a node edge structure and a feature association graph. The overall joint state probability of the feature nodes is calculated. The joint state probability is the joint occurrence probability of multiple feature nodes in certain states at the same time. The linkage effect between features is quantified to obtain the feature conditional dependency analysis results.

[0061] The spatial feature distribution analysis module performs a detailed analysis of the distribution of each feature node in the multidimensional feature space based on the feature condition dependency analysis results. By comparing the probability of a single feature node with the overall joint state probability of the feature nodes, the probability of a single feature node is the probability of being in a certain state independently without considering other nodes. The synergy of the feature combination is judged through a preset ratio threshold. By calculating the synergy score of the feature combination, the contribution of adjacent nodes and connecting edges in the feature combination is quantified, and the feature nodes and edges that contribute to the overall dependency are identified and divided into influential feature combinations. The node and edge information of the influential feature combination is extracted to construct a feature cluster dependency structure.

[0062] The health risk prediction module dynamically monitors the changes of each feature cluster in the time series based on the data of the feature cluster dependency structure, divides the heart data into fixed windows by time, analyzes the state changes of the feature cluster in each time window, quantifies the fluctuation of the feature cluster by fuzzy entropy value, and obtains the fluctuation index of the feature cluster. The stability of the health status is judged according to the fluctuation index, and the individual's health status is divided into three risk levels: low, medium and high, to generate the health risk assessment results;

[0063] Based on the health risk assessment results, the risk group identification module performs cluster analysis on individual data points that are judged to be at high risk levels, identifies high-risk feature clusters, recursively analyzes the neighborhood density of high-risk feature clusters, identifies the areas with the highest density, combines the features of high-risk clusters to form a group feature description information set, and generates risk group identification results in heart disease data; it is used to provide real-time monitoring suggestions and early warning analysis for high-risk groups, as well as data to assist in decision-making support;

[0064] The results of feature conditional dependency analysis include key feature nodes of heart disease, such as age node, blood pressure node, cholesterol node and heart rate node, as well as the highly dependent edge structure between these nodes. The feature cluster dependency structure includes the edges of core feature nodes such as hypertension and high cholesterol, as well as other nodes and edges that contribute highly to the development of heart disease. The health risk assessment results include the description of the feature matrix state of each time window, the volatility index score of the feature cluster, and the health status risk level divided by this score. The results of risk group identification in heart disease data include the feature cluster description of the highest density area, the group feature information set, and the potential high-risk status of these groups.

[0065] See also Figure 2 ,The specific steps for obtaining the degree of conditional correlation between feature nodes are:

[0066] Based on the collected heart data, the characteristics of heart disease including age, blood pressure, cholesterol, and heart rate information are extracted and preprocessed, and the processed data are mapped into the multidimensional feature space as feature nodes. Each node contains its feature value and corresponding sample index, and the feature node data set is obtained;

[0067] Based on the collected heart data, the characteristics of heart disease including age, blood pressure, cholesterol, and heart rate information are extracted and preprocessed. The collected heart data are preliminarily screened by using data cleaning methods. Missing value processing techniques such as mean filling or interpolation methods are used to repair data gaps. Then outlier detection is performed. Outliers outside the normal range are removed using box plots or standard deviation-based outlier detection. In blood pressure data processing, blood pressure readings are converted to standardized values ​​and are unitless through Zscore standardization for comparison. In cholesterol and heart rate data preprocessing, the MinMax normalization method is used to map the data to the [0,1] interval to ensure that the scales of different types of features are unified. Then, principal component analysis (PCA) is used to reduce the dimension of the feature dimension to extract the main information and reduce redundancy. These processed data are mapped as feature nodes into the multidimensional feature space. Each node contains its feature value and the corresponding sample index. For example, the original data of a sample is age 45 years old, blood pressure 135 / 85 mmHg, cholesterol 240 mg / dL, and heart rate 72 bpm. After preprocessing, age data is normalized to the range of [0,1], for example, it is mapped to 0.5; blood pressure is represented as 0.3 after Zscore normalization; cholesterol is 0.7 after MinMax normalization, and heart rate is mapped to 0.4. Finally, the data of this sample can be represented as a feature vector These feature vectors are organized into high-dimensional data points. In the multidimensional feature space, each data point represents a feature combination of a sample. This process maps all samples into the multidimensional space, making each data point a feature node in the space. For example, for 1,000 samples, each sample will occupy a different coordinate position, and the feature nodes are formed according to their respective feature vector combinations. The nodes are used to analyze the dependencies and distribution characteristics between the features in the data set.

[0068] According to the characteristic node data set, the formula is used:

[0069] ;

[0070] Calculated at feature nodes In the event of a The conditional probability , get the analysis results of the correlation between adjacent feature nodes;

[0071] in, is the adjacent feature node and The joint probability of simultaneous occurrence is recorded by analyzing the sample data row by row. and The number of samples that meet specific conditions at the same time is divided by the total number of samples to obtain It is a feature node The marginal probability of The frequency of occurrence in the sample is calculated by statistically analyzing the Get the number of occurrences and divide it by the total number of samples.

[0072] If in a heart disease dataset containing 500 samples, the feature Indicates high cholesterol level (>240 mg / dL), characteristic Indicates high blood pressure (>140 / 90mmHg). There are 80 data in the sample that meet the conditions of high cholesterol and high blood pressure at the same time. Then:

[0073] ;

[0074] There are 200 data with hypertension in the sample, then:

[0075] ;

[0076] Calculate the conditional probability:

[0077] ;

[0078] This result shows that in the sample, given high blood pressure, the conditional probability of high cholesterol is 0.4. According to the calculation results The degree of conditional correlation between feature nodes can be analyzed, as follows: The value is close to (Feature Node ) indicates that the node and nodes The correlation between them is weak, that is, at the node Under the conditions that have occurred, the node The probability of occurrence does not change significantly. On the contrary, if The value is significantly higher or lower than , it indicates that the node The occurrence of the node have a stronger influence and show higher conditional correlation. For example, in a heart disease dataset: Indicates high cholesterol (>240 mg / dL). represents high blood pressure (>140 / 90 mmHg). In the sample, the node The marginal probability is 0.3 (i.e. 30% of the data in the sample show high cholesterol). is 0.4 (i.e., if hypertension has already occurred, the probability of high cholesterol is 40%). and .because Greater than , it can be seen that when the node When (hypertension) occurs, the node The probability of (high cholesterol) occurring increases. This indicates that there is a certain positive correlation between high blood pressure and high cholesterol.

[0079] See also Figure 3 ,The specific steps for obtaining the linkage effect between quantitative features are:

[0080] According to the analysis results of the correlation between adjacent feature nodes, the connection weight between nodes is determined. If the weight exceeds the preset correlation threshold, a connection edge is established between the nodes, and the weight value is attached to the edge and recorded in the adjacency matrix or adjacency table to generate a feature association graph.

[0081] According to the calculated , which can be used to connect feature nodes to each other. The process of constructing a feature association graph is as follows: First, according to The value of determines the connection weight between nodes. If the preset correlation threshold is exceeded, and nodes An edge is established between them, and this weight value is attached to the edge. Each pair of nodes that meet the conditions is checked for relevance and connected by edges in turn. The information of nodes and edges is recorded in an adjacency matrix or adjacency table, where the matrix elements or table entries represent the weight values ​​between the corresponding node pairs. For example, suppose the feature node indicates high cholesterol, node Indicates high blood pressure, , if the correlation threshold is set to 0.3, then at node and nodes An edge is established between them with a weight of 0.4. Similarly, if there is another node (e.g. high heart rate) with nodes Relevance , and this value also exceeds the correlation threshold of 0.3, then at the node and nodes An edge is established between them with a weight of 0.5. Finally, the nodes and the connected edges are plotted in the graph so that each node can be connected to other nodes that meet the conditions to form a complete network structure. The structure of the graph represents the dependency between features, and the edge weights in the graph show the strength of the correlation between nodes. This information can be further stored using the adjacency matrix in graph theory. In this way, the conditional correlation between all feature nodes is systematically represented as a feature association graph.

[0082] According to the node and edge information in the feature association graph, the formula is used:

[0083] ;

[0084] Compute feature nodes , , The joint probability of being in certain states at the same time , get the feature condition dependency analysis results;

[0085] in, is the conditional probability, indicating that at the node and Under the conditions that have occurred, the node The probability of occurrence is obtained by filtering out the data set that satisfies and All samples of the state, and count those that satisfy The number of samples of the state, record the number of samples that meet the conditions and divide them to meet and The total number of samples under each condition is used to obtain the conditional probability value. is the conditional probability, indicating that at the node Under the conditions that have occurred, the node The probability of occurrence is obtained by filtering all The samples of the state and count those with The number of samples of the state is then divided by the number of samples to satisfy The total number of samples of the state to obtain the conditional probability value, is the edge probability, indicating that the node The probability of appearing in the sample data is calculated by statistically satisfying the node The marginal probability is obtained by dividing the number of samples of the state by the total number of samples.

[0086] For example, in a heart disease dataset containing 500 samples: Represents high cholesterol (>240mg / dL), characteristics Represents high blood pressure (>140 / 90 mmHg), characteristics Represents a high heart rate (>100bpm). In statistics, the following data is obtained: The frequency of occurrence is 120 times:

[0087] ;

[0088] In cases with high heart rate ( ) in the case of hypertension ( ) appears 60 times:

[0089] ;

[0090] In the presence of high heart rate and high blood pressure, high cholesterol ( ) appears 40 times:

[0091] ;

[0092] Plugging these values ​​into the formula to calculate the overall joint probability:

[0093] ;

[0094] ;

[0095] The results showed that the joint probability of high cholesterol, high blood pressure and high heart rate occurring simultaneously was 0.0804, or 8.04%.

[0096] According to the calculation results To quantify the linkage effect between features, it is necessary to compare and analyze the joint probability with the marginal probability and conditional probability of each feature, as follows:

[0097] Calculate the marginal probability of each feature, such as , , These probabilities reflect the frequency with which each feature occurs individually. The marginal probabilities can be obtained by counting the number of times each feature appears in the sample data and dividing it by the total number of samples.

[0098] The joint probability indicates the probability of multiple feature nodes appearing at the same time. The larger its value, the more common the simultaneous appearance of these feature combinations is in the sample. , and It is calculated by dividing the number of samples that appear at the same time by the total number of samples.

[0099] The joint probability The product of the marginal probabilities when assuming independence of the features For comparison. , it indicates that there is a positive correlation and linkage effect between these features, that is, in actual data, the possibility of these features appearing simultaneously is greater than the possibility of them appearing independently. , it means that the correlation between these features is weak. , it means that there may be a negative correlation between the features, that is, the probability of their joint occurrence is lower than expected when they are independent.

[0100] By defining the linkage coefficient :

[0101] ;

[0102] when , indicating that there is a positive linkage effect between features. , indicating that there is no significant linkage effect and the features are basically independent. , indicating that there is a negative linkage effect between features.

[0103] For example, in a heart disease dataset containing 500 samples: (Probability of high cholesterol) (Probability of high blood pressure) (Probability of high heart rate) (joint probability);

[0104] Calculate the linkage coefficient :

[0105] ;

[0106] because , which means that the frequency of high cholesterol, high blood pressure and high heart rate occurring together in the sample is higher than the expected value when they are independent, indicating that there is a positive linkage effect between them.

[0107] See also Figure 4 , the specific steps for obtaining the quantified contribution of adjacent nodes and connecting edges in the feature combination are:

[0108] Based on the feature condition dependency analysis results, by comparing the edge probability of a single feature node with the probability in the joint state, the feature combination with synergistic effect is identified and marked, and the feature combination synergistic effect analysis results are obtained;

[0109] The distribution of each feature node in the multidimensional feature space is analyzed in detail. For example, in the heart disease data set, the feature node (High cholesterol), (High Blood Pressure) and (high heart rate) and their respective marginal probabilities are , , The probability of the joint state The ratio of the probability combination of a single feature node (i.e., the product of the edge probabilities, 0.3×0.4×0.2=0.024) to the probability in the joint state is calculated, and the result is The ratio thresholds are set to 1.2 and 0.8 respectively. If the ratio is greater than 1.2, it indicates that there is a positive synergistic effect in the feature combination, that is, the actual frequency of the three feature combinations is higher than the expected value in the independent case. If the ratio is lower than 0.8, it indicates a negative synergistic effect, that is, the actual frequency of the combination is lower than the expected value. Through this ratio comparison process, feature combinations with synergistic effects can be identified and marked for further analysis and recording.

[0110] According to the analysis results of the synergistic effect of feature combination, the formula is adopted:

[0111] ;

[0112] Calculating the synergy score , get the overall synergistic effect data of the feature combination;

[0113] in, It is a feature node The weight of the node indicates the relative importance of the node in the feature combination. The weight can be analyzed through historical data. For example, when analyzing past heart disease data, it is found that hypertension (node ) has a greater impact on the occurrence of the disease, and its weight will be relatively high. is the conditional probability, indicating the feature node On its adjacent nodes The probability of occurrence under given conditions is calculated by counting the number of samples that meet the conditions. Under the conditions The frequency of occurrence is obtained, It is a representation and feature node A set of adjacent feature nodes is determined by analyzing the feature dependency graph or adjacency matrix to determine which feature nodes are adjacent to each other. Direct connection, It is a feature node The probability of independent occurrence in the data set, through the statistical node The number of times it appears in the sample is divided by the total number of samples. Used to accumulate the probability of independent occurrence of each feature node , is the total number of feature nodes involved in the feature combination.

[0114] For example, in a heart disease prediction dataset containing 500 samples, there are three feature nodes , and , representing high cholesterol (>240mg / dL), high blood pressure (>140 / 90mmHg) and high heart rate (>100bpm), respectively, with weights: , , , conditional probability: , , , marginal probability: , , .

[0115] Calculate the synergy score:

[0116] ;

[0117] ;

[0118] The results showed that the synergistic effect score was 0.787.

[0119] See also Figure 5 , the specific steps for obtaining the feature cluster dependency structure are:

[0120] According to the overall synergy effect data of the feature combination, the synergy effect score is compared with the preset scoring standard, the feature combination with a score higher than the standard is screened, the corresponding node and edge information is extracted, and the influential feature combination is obtained;

[0121] Identify the feature nodes and edges that contribute to the overall dependency and score them based on the calculated synergy effect , compare the score with the preset score standard, for example, the preset score standard is 0.75, extract the feature combinations with scores higher than the standard through the screening step, use data processing tools (such as the pandas library in Python) to screen and sort, and calculate the The feature combination with a value of 0.787 was compared with the preset standard, and the result showed that the score was higher than 0.75. Therefore, the feature combination was marked as contributing to the overall dependency. Then the information of related nodes and edges was extracted, and the conditional probability values ​​of these nodes and the relative positions in the combination were checked in detail to confirm the connection strength and correlation between the nodes. By recording the nodes and edges of the high-scoring feature combinations, the influential feature combinations were finally divided.

[0122] The information of nodes and edges in the influential feature combination is recorded in the adjacency matrix or adjacency table according to the weight, and the nodes and edges are added in sequence, the weight value of each edge is marked, and the related nodes are grouped and merged to establish the feature cluster dependency structure;

[0123] By selecting the node and edge information of influential feature combinations, first extract the feature combination node and edge data with scores higher than the preset standard from the previous analysis results, use the adjacency matrix or adjacency table to represent the connection relationship between nodes and edges, take each feature node as the vertex of the graph, assign weights to the connection relationship between nodes according to the contribution score, and record the weight information of nodes and edges to ensure that the real dependency relationship of the feature combination is reflected. Next, apply clustering algorithms such as hierarchical clustering to automatically group nodes with high correlation into the same cluster. The hierarchical clustering process is performed by referring to the distance or similarity matrix between nodes, and gradually merges the most similar nodes and edges. During the construction process, add node and edge information to the structure one by one, annotate the weight value of each edge, and verify whether the nodes of each cluster meet the predetermined dependency and contribution. By checking the node connection in each cluster, ensure that the edge weight and contribution between nodes are consistent with the previous analysis. Then identify the central node in each cluster. The central node represents the feature with the strongest dependency in the cluster, perform structural consistency check, correct and optimize the arrangement of nodes and edges to reflect the actual dependency and contribution between features, and form a complete feature cluster dependency structure.

[0124] See also Figure 6 , the specific steps for obtaining the volatility index of the feature cluster are:

[0125] Based on the data of feature cluster dependency structure, the cardiac data is divided into fixed time windows according to time, and the node state information in each feature cluster is extracted in each time window. The state changes in adjacent time windows are compared, the fluctuation of feature clusters is marked, and the state change information of feature clusters in the time window is obtained;

[0126] Each feature cluster is dynamically monitored according to the time series, and the cardiac data is divided into several sections according to fixed time windows, so as to analyze the state changes of the feature cluster in detail in each time window. First, the initial state of each feature node in the feature cluster is recorded in the first time window, and then the state information of each feature node is extracted in turn in each time window, including the numerical change amplitude of the node, state offset, etc. On this basis, by comparing the numerical changes of the feature clusters in adjacent time windows, the fluctuation amplitude of each feature node is marked, and the state changes of the feature clusters in each time period are sorted out in turn, and the fluctuation curve of the feature cluster in each time window is gradually generated. During the analysis process, the state data of the feature cluster in each time window will be gradually updated in chronological order, so that the dynamic changes of the feature cluster can be fully presented in the time series, and finally the fluctuation of each feature cluster in the time series is obtained.

[0127] According to the feature cluster state change information within the time window, the formula is used:

[0128] ;

[0129] Calculate fuzzy entropy value , get the volatility index of the characteristic cluster;

[0130] in, Indicates the degree of disorder of feature clusters within a specific time window, including the influence of multi-dimensional feature node interactions, and is the main indicator for measuring the fluctuation of feature clusters. is the index variable of the feature node, used to indicate the feature nodes, is within the time window The similarity probability of feature nodes can be calculated by counting the frequency of occurrence of feature nodes in a certain time window and normalizing them into probability values. is the number of main feature nodes in the feature cluster, usually representing the number of features in heart health data (such as age, blood pressure, cholesterol, etc.), is the weight adjustment coefficient, which is used to adjust the sensitivity of the feature cluster fluctuation index. It is usually determined by the changes in the health monitoring data. If a feature (such as heart rate) fluctuates slightly under normal conditions, but fluctuates significantly under abnormal conditions, a higher weight can be given to the heart rate to more sensitively reflect the impact of heart rate changes on health status. By comparing the fluctuation range of normal and abnormal data, analyzing the sensitivity of each feature to changes in health status, and adjusting the weight coefficient accordingly, the individual's health status can be more accurately reflected. is the index variable of the feature node, indicating the The number of non-similar feature nodes, It is The non-similarity probability of feature nodes represents the feature node The probability of appearing in a different state from other nodes within the time window is used to distinguish the fluctuation of feature clusters. It can be obtained by normalizing the frequency of statistical node changes exceeding the threshold. It is the number of non-similar feature nodes in the feature cluster, that is, the total number of non-similar nodes. By screening the feature nodes in the feature cluster, the nodes that do not meet the similarity conditions can be classified as non-similar nodes.

[0131] If the feature cluster contains 3 main nodes (similar nodes) and 2 auxiliary nodes (non-similar nodes), the similarity probability is , , , the probability of non-similarity is , ,and .

[0132] Substitute the formula to calculate the fuzzy entropy:

[0133] ;

[0134] The results show that the fuzzy entropy value of the feature cluster in this time window is 0.7523.

[0135] Convert the fuzzy entropy value into a fluctuation index of feature clusters:

[0136] The numerical range of the fuzzy entropy value calculated usually fluctuates greatly, and the fuzzy entropy value of each feature cluster may be different, so the fuzzy entropy value is normalized first. The maximum value normalization method can be used to normalize the fuzzy entropy value to the range of 0 to 1.

[0137] For example, the normalization formula can be:

[0138] ;

[0139] in is the fuzzy entropy value of the current feature cluster, and They are the minimum and maximum fuzzy entropy values ​​in the sample data respectively. The fuzzy entropy value is made suitable for use in the feature cluster fluctuation indicator through normalization.

[0140] Define a fluctuation threshold based on the fluctuation characteristics and normal range of health data , which is used to distinguish low volatility from high volatility. The analysis can be based on the fuzzy entropy distribution of historical health data, and the median or mean of the data can be selected as the initial threshold. In this way, the volatility can be mapped to an indicator for analysis.

[0141] The normalized fuzzy entropy value Fluctuation threshold 'To determine the volatility level. The volatility index of the characteristic cluster can be divided into three levels: low, medium and high: , then the volatility level is low volatility. , then the volatility level is medium volatility. , the volatility level is high volatility.

[0142] Map the fluctuation level to the value to get the fluctuation index of the feature cluster. For example, low fluctuation can correspond to 0.2, medium fluctuation corresponds to 0.5, and high fluctuation corresponds to 1. In this way, a quantitative fluctuation index can be obtained for subsequent health status analysis.

[0143] See also Figure 7 , the specific steps for obtaining health risk assessment results are:

[0144] According to the fluctuation index of the feature cluster, by comparing the fluctuation index in each time window with the stability threshold, the fluctuation amplitude of the feature cluster is determined, and the change trend of the fluctuation index in the time window is monitored to obtain the stability analysis result of the health state;

[0145] Based on volatility indicators To judge the stability of health status, the normalized fuzzy entropy value is used Based on the individual fluctuations, the state is judged. For example, if a feature cluster is calculated within a time window , and the stability threshold is set to 0.5, then the fluctuation index is significantly lower than the threshold, indicating that the fluctuation amplitude of the feature cluster is small and belongs to the range of healthy and stable state. If the value is close to or exceeds 0.5, especially above 0.6, it indicates that the characteristic cluster has large fluctuations, which may indicate unstable health status. In addition, the trend change of the fluctuation index over time is also used to further judge health stability: for example, if A gradual increase in multiple time windows, even if the value in a single window does not exceed the threshold, may indicate potential health status fluctuations. Therefore, the dynamic stability of an individual's overall health status can be judged by the gradually accumulated fluctuation index.

[0146] According to the stability analysis results of the health status, the fluctuation indicators in each time window are divided into multiple risk levels, the risk level information of each time window is recorded, and the health risk assessment results are generated;

[0147] The individual's health status is divided into three risk levels: low, medium and high. The numerical range of defines the classification criteria for each level. Specifically, if If it is less than 0.4, the health status is classified as low risk; if Between 0.4 and 0.6, it is defined as medium risk level; When it exceeds 0.6, it indicates that the individual is at a high risk level. The fluctuation indicators in each time window can be classified by these levels, making the health risk level a reference data reflecting the individual's current health status. By continuously monitoring the risk level in different time windows, the trend changes in the health risk level can be identified, and finally the health risk assessment results can be generated, providing a basis for subsequent health management and heart disease risk prevention and control.

[0148] See also Figure 8 , the specific steps for obtaining the risk group identification results in heart disease data are:

[0149] Based on the health risk assessment results, extract individual data points that are judged to be high-risk, identify high-risk feature clusters, organize the value range, fluctuations and abnormal points of each feature cluster into a data set, and screen them to obtain risk feature cluster information;

[0150] Cluster analysis is performed on individual data points judged to be at a high risk level. High-risk individuals are extracted from the assessment results, and their feature clusters are identified one by one. Their feature information is organized into a usable data set. The high-risk feature clusters are screened, and the numerical range, fluctuation and abnormal point marking of each feature cluster are used as the basic data for cluster analysis. By clustering and classifying the neighborhood features of high-risk data points, the neighborhood features with similar or overlapping features with these high-risk individuals are marked as potential risk clusters, and finally the risk feature cluster information is identified.

[0151] According to the risk feature cluster information, the formula is used:

[0152] ;

[0153] Calculate neighborhood density , identifying the areas with the highest density and generating risk group identification results in heart disease data;

[0154] in, Indicates the density of high-risk feature clusters in a specific space and is used to quantify the concentration of feature clusters. It is an index variable, indicating the number of a specific feature cluster, which represents the data points and volume of each high-risk feature cluster in turn, helping to compare and accumulate the density values ​​of high-risk clusters one by one. It is The number of high-risk data points in a feature cluster is used to represent the total number of risk feature points in the feature cluster. It is obtained by counting the feature data points that meet the high-risk conditions in the neighborhood. It is The neighborhood volume of feature clusters represents the volume occupied by each feature cluster in space. The distance calculation method is used to estimate the distribution area or volume of each feature cluster in space. The clustering algorithm is applied to measure the volume. The neighborhood in space is divided into units of the same or different volumes. The volume of each feature cluster is generated through the neighborhood distribution. is the number of high-risk feature clusters in the neighborhood, indicating the total number of feature clusters that meet the high-risk criteria in a specific neighborhood.

[0155] If a neighborhood contains three high-risk feature clusters, the number and volume of high-risk data points in each feature cluster in the neighborhood are: , ; , ; , .

[0156] Substitute the formula to calculate the neighborhood density :

[0157] ;

[0158] ;

[0159] ;

[0160] This result shows that the neighborhood density , and compare this value with the preset density threshold. The density threshold can be set based on the density range of known high-risk areas in historical data, or by setting an upper quantile of density distribution (such as the 75% quantile) as the high-density boundary through statistical analysis. If the density threshold is exceeded, the area is considered to be a high-density area, that is, an area where high-risk data points are highly clustered in space. By comparing with the threshold, the areas with density higher than the threshold are screened out, and these areas are identified as concentrated areas with a higher risk of heart disease. After identifying the high-density areas, the features in these high-risk feature clusters are further analyzed, including key health indicators such as age, blood pressure, and cholesterol levels. These features are aggregated to form a group feature set, which represents the common health features in the high-density area. The similarity features in the feature combination are extracted and the numerical range of the features is recorded, such as the common features in age, heart rate, etc., and the fluctuation range of these indicators. These feature combination data are integrated into a descriptive information set to identify the typical health features in the high-risk area. The group feature description information set is applied to other data points in the heart disease data, and groups with the same or similar features are screened out by matching the similarity of feature combinations, thereby identifying potential risk groups. The generated risk group identification results contain group characteristic information, number of data points and spatial distribution of each high-risk area, which are used for further monitoring and early warning analysis of potential high-risk groups for heart disease, and provide data reference for doctors and decision support systems.

[0161] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. AI-based heart disease prediction system, characterized by: The system comprises: The feature relationship modeling module maps each type of information into a multidimensional feature space based on the collected cardiac data, analyzes the degree of conditional correlation between feature nodes, connects feature nodes to each other through correlation information, quantifies the linkage effect between features, and obtains the feature conditional dependency analysis results; The spatial feature distribution analysis module performs a detailed analysis on the distribution of feature nodes in the multidimensional feature space based on the feature condition dependency analysis results, determines the synergy of feature combinations, quantifies the contribution of adjacent nodes and connecting edges in the feature combinations, divides them into influential feature combinations, and constructs a feature cluster dependency structure; The health risk prediction module divides the heart data into fixed windows by time based on the data of the feature cluster dependency structure, analyzes the state changes of the feature clusters in each time window, quantifies the fluctuation of the feature clusters, calculates the fuzzy entropy value, converts the fuzzy entropy value into a fluctuation index of the feature cluster, judges the stability of the health state according to the fluctuation index, and generates a health risk assessment result; Based on the health risk assessment results, the risk group identification module performs cluster analysis on individual data points determined to be at a high risk level, identifies feature clusters with risks, and generates risk group identification results in heart disease data.

2. The AI-based heart disease prediction system according to claim 1, characterized in that: The step of obtaining the conditional correlation degree between the analysis feature nodes is specifically as follows: Based on the collected heart data, the characteristics of heart disease including age, blood pressure, cholesterol, and heart rate information are extracted and preprocessed, and the processed data are mapped into the multidimensional feature space as feature nodes. Each node contains its feature value and corresponding sample index, and the feature node data set is obtained; According to the characteristic node data set, the formula is adopted: ; Calculated at feature nodes In the event of a The conditional probability , get the analysis results of the correlation between adjacent feature nodes; in, is the adjacent feature node and The joint probability of simultaneous occurrence, It is a feature node The marginal probability of The frequency of occurrence in the sample.

3. The AI-based heart disease prediction system according to claim 2, characterized in that: The steps for obtaining the linkage effect between the quantitative features are specifically as follows: According to the analysis result of the correlation degree of the adjacent feature nodes, the connection weight between the nodes is determined. If the weight exceeds the preset correlation threshold, a connection edge is established between the nodes, and the weight value is attached to the edge and recorded in the adjacency matrix or adjacency table to generate a feature association graph; According to the node and edge information in the feature association graph, the formula is adopted: ； Compute feature nodes , , The joint probability of being in certain states at the same time , get the feature condition dependency analysis results; in, is the conditional probability, indicating that at the node and Under the conditions that have occurred, the node The probability of occurrence, is the conditional probability, indicating that at the node Under the conditions that have occurred, the node The probability of occurrence, is the edge probability, indicating that the node The probability of appearing in the sample data.

4. The AI-based heart disease prediction system according to claim 3, characterized in that: The specific steps of obtaining quantified contributions of adjacent nodes and connecting edges in the feature combination are: Based on the feature condition dependency analysis result, by comparing the edge probability of a single feature node with the probability in a joint state, identifying and marking feature combinations with synergistic effects, and obtaining feature combination synergistic effect analysis results; According to the analysis results of the synergistic effect of the feature combination, the formula is adopted: ; Calculating the synergy score , get the overall synergistic effect data of the feature combination; in, It is a feature node The weight of is the conditional probability, indicating the feature node On its adjacent nodes The probability of occurrence under given conditions, It is a representation and feature node The set of adjacent feature nodes, It is a feature node The probability of appearing independently in the data set, Used to accumulate the probability of independent occurrence of each feature node , is the total number of feature nodes involved in the feature combination.

5. The AI-based heart disease prediction system according to claim 4, characterized in that: The steps for obtaining the feature cluster dependency structure are specifically as follows: According to the overall synergistic effect data of the feature combination, the synergistic effect score is compared with the preset scoring standard, the feature combination with a score higher than the standard is screened, the corresponding node and edge information is extracted, and the influential feature combination is obtained; The information of nodes and edges in the influential feature combination is recorded in an adjacency matrix or an adjacency table according to the weights, nodes and edges are added in sequence, the weight value of each edge is marked, the associated nodes are grouped and merged, and a feature cluster dependency structure is established.

6. The AI-based heart disease prediction system according to claim 5, characterized in that: The steps for obtaining the volatility index of the characteristic cluster are specifically as follows: Based on the data of the feature cluster dependency structure, the cardiac data is divided into fixed time windows by time, the node state information in each feature cluster is extracted in each time window, the state changes in adjacent time windows are compared, the fluctuation of the feature cluster is marked, and the state change information of the feature cluster in the time window is obtained; According to the feature cluster state change information within the time window, the formula is used: ; Calculate fuzzy entropy value , get the volatility index of the characteristic cluster; in, is the index variable of the feature node, is within the time window The similarity probability of feature nodes, is the number of main feature nodes in the feature cluster, is the weight adjustment coefficient, is the index variable of the feature node, It is The dissimilarity probability of feature nodes, is the number of non-similar feature nodes in the feature cluster.

7. The AI-based heart disease prediction system according to claim 6, characterized in that: The steps for obtaining the health risk assessment results are specifically as follows: According to the fluctuation index of the characteristic cluster, by comparing the fluctuation index in each time window with the stability threshold, the fluctuation amplitude of the characteristic cluster is determined, and the change trend of the fluctuation index in the time window is monitored to obtain the stability analysis result of the health state; According to the stability analysis result of the health status, the fluctuation index in each time window is divided into multiple risk levels, the risk level information of each time window is recorded, and the health risk assessment result is generated.

8. The AI-based heart disease prediction system according to claim 7, characterized in that: The steps for obtaining the risk group identification result in the heart disease data are specifically as follows: Based on the health risk assessment results, extract individual data points determined to be high risk, identify high-risk feature clusters, organize the value range, fluctuation and abnormal points of each feature cluster into a data set, and screen them to obtain risk feature cluster information; According to the risk feature cluster information, the formula is used: ; Calculate neighborhood density , identifying the areas with the highest density and generating risk group identification results in heart disease data; in, represents the number of the feature cluster, It is The number of high-risk data points in a feature cluster, It is The neighborhood volume of feature clusters, is the number of high-risk feature clusters in the neighborhood.

Images