Medical data analysis method and system based on block chain and privacy protection

Abstract

The invention relates to the technical field of medical data security, in particular to a medical data analysis method and system based on a block chain and privacy protection, and the method comprises the following steps: analyzing the correlation degree between nodes through the correlation between sample features according to the local medical data of a medical institution, taking data samples as the nodes of a graph, and taking the data samples as the nodes of the graph; and analyzing the global homogeneity of the sub-graphs, and generating a cross-mechanism medical data association graph structure. According to the method, medical data nodes and relationships thereof are analyzed in multiple dimensions, privacy protection and conjoint analysis efficiency is optimized, data sample implicit association is accurately captured, a global view angle is enhanced, redundant information interference is reduced through association degree matrix, high-sensitivity data nodes are eliminated, and disturbance processing is performed on low-sensitivity nodes; the privacy protection intensity is remarkably improved, the analysis value is reserved, the data association accuracy is optimized through the dynamically-adjusted node edge relation, stronger feature matching is achieved, and the joint analysis capacity under the privacy protection condition is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of medical data security technology, and in particular to a medical data analysis method and system based on blockchain and privacy protection. Background Art

[0002] The field of medical data security technology includes research and application in the protection, management, and analysis of medical data, aiming to ensure the privacy, integrity, and availability of medical data. With the rapid development of information technology, the digitization and networking of medical data are becoming increasingly popular. How to ensure the security of sensitive data during storage, transmission, and processing has become a core issue in this technology field. Medical data security technology involves a variety of technical means such as data encryption, access control, identity authentication, data isolation, and privacy protection algorithms to ensure the security and compliance of medical data in multi-party use. With the rise of artificial intelligence and big data technologies, how to achieve the sharing and effective analysis of medical data while ensuring data privacy has also become an important research topic in this field.

[0003] Among them, the medical data privacy protection and analysis method refers to a method of protecting medical data privacy and performing data analysis through a technical framework based on federated learning. The patent subject involves the use of federated learning technology to achieve distributed data processing, enabling different medical institutions to jointly train machine learning models without sharing original data. To ensure data privacy, this patent subject also combines differential privacy technology to add noise during the data processing process to prevent the leakage of sensitive information. The patent also proposes a multi-level security strategy to ensure that all participants follow strict privacy protection rules during the data analysis process. The patent content includes technical measures such as the construction of a federated learning framework, the design of a privacy protection algorithm, a data exchange mechanism between participants, and a data encryption and decryption solution.

[0004] Existing technologies have certain limitations in ensuring medical data privacy protection and joint analysis. Since medical data is stored in an isolated manner, cross-institutional data sharing relies on direct data transmission or centralized storage. This model easily leads to the risk of sensitive information being leaked during transmission. Even with the use of encryption and differential privacy technologies, the compromise in data analysis accuracy leads to a decline in the performance of the trained model. The node weight screening and data perturbation methods of existing technologies are only based on simple rules and lack in-depth mining of global data relationships. It is difficult to effectively eliminate highly sensitive nodes, resulting in potential privacy leakage risks. Existing technologies lack dynamic adjustment and feature fusion mechanisms in cross-institutional data collaboration, resulting in the analysis results being difficult to fully reflect the actual data correlation and distribution patterns, resulting in the inability to achieve a balance between medical data privacy protection and analysis performance, and it is difficult to meet the high privacy and efficiency requirements in practical applications. Summary of the Invention

[0005] In order to solve the limitations of existing technologies in ensuring medical data privacy protection and joint analysis, since medical data is stored in an isolated manner, cross-institutional data sharing relies on direct data transmission or centralized storage. This model easily leads to the risk of sensitive information being leaked during transmission. Even if encryption and differential privacy technologies are used, the compromise in data analysis accuracy leads to a decline in the performance of the trained model. The node weight screening and data perturbation methods of the existing technology are only based on simple rules, lacking in-depth mining of global data relationships, making it difficult to effectively eliminate highly sensitive nodes, resulting in privacy leakage risks. The existing technology lacks dynamic adjustment and feature fusion mechanisms in cross-institutional data collaboration, resulting in the analysis results being difficult to fully reflect the real data correlation and distribution patterns, resulting in the inability to achieve a balance between medical data privacy protection and analysis performance, and the difficulty in meeting the high privacy and efficiency requirements in practical applications. The embodiment of the present invention provides a medical data analysis method and system based on blockchain and privacy protection. The technical solution is as follows:

[0006] In one aspect, a medical data analysis method based on blockchain and privacy protection is provided, comprising:

[0007] S1: Based on the local medical data of medical institutions, the correlation between nodes is analyzed through the correlation between sample features. The data samples are used as nodes of the graph. The global homogeneity of the subgraphs is analyzed in the same way to generate the cross-institutional medical data association graph structure.

[0008] S2: Analyze the connection strength of the nodes in the global graph based on the node and edge information in the cross-institutional medical data association graph structure, combine the connection strength data with the edge weight distribution parameters to filter the node weights, eliminate the associated node information according to the threshold range, and generate a screening result of low-sensitivity nodes;

[0009] S3: Based on the screening results of the low-sensitivity nodes, a disturbance amplitude analysis is performed according to the edge weight distribution parameters, and the node weight matrix is ​​reconstructed after adding the noise value based on the local subgraph weight ratio to generate a perturbed low-sensitivity node weight matrix;

[0010] S4: Using the perturbed low-sensitivity node weight matrix, extracting node feature data in the matrix, analyzing the correlation degree of node embedding values ​​through homogeneity comparison, recombining the correlation value and edge weight into weighted features, dynamically adjusting the node and edge relationship in combination with weight offset, and generating a feature-fused node-edge relationship matrix;

[0011] S5: Based on the node-edge relationship matrix of the feature fusion, extract the node feature distribution values ​​and edge weight distribution in the matrix, iteratively adjust the weight parameters, statistically distribute and integrate the weight data, and generate the medical data privacy protection joint analysis distribution results.

[0012] As a further solution of the present invention, the cross-institutional medical data association graph structure includes node information, edge information, and a global homogeneity matrix. The screening results of the low-sensitive nodes include the node weights after screening, the edge weight distribution parameters within the screening range, and the sensitivity score range. The perturbed low-sensitive node weight matrix includes the perturbed node weight values, the node weight ratio parameters, and the perturbed position mapping matrix. The feature-fused node-edge relationship matrix includes the node feature distribution values, the edge weight distribution, and the dynamically adjusted weight features. The medical data privacy protection joint analysis distribution results include the adjusted weight parameter distribution, the global node relationship statistical distribution, and the global weight integration results.

[0013] As a further solution of the present invention, based on the local medical data of medical institutions, the correlation between nodes is analyzed by correlation between sample features, the data samples are used as nodes of the graph, and the global homogeneity of the subgraphs is analyzed by homogeneity. The specific steps of generating a cross-institutional medical data association graph structure are as follows:

[0014] S101: Based on the local medical data of the medical institution, the data samples are parsed into independent nodes, the statistics of the numerical features are extracted, the value distribution of the categorical features is extracted, the feature interaction parameters are analyzed, the node association strength is filled into the matrix, and the sample association feature matrix is ​​generated;

[0015] S102: Based on the sample association feature matrix, convert the node association strength values ​​in the matrix according to a normalized range, extract the node adjacency relationship, group the nodes according to the node adjacency parameters, assign the nodes in the group to subgraphs, adjust the subgraph node layout, and generate a sample subgraph structure model;

[0016] S103: Based on the sample subgraph structure model, the node matching relationship between subgraphs is analyzed, the cross-subgraph node connection parameters and association strength values ​​are calculated, the global subgraph homogeneity values ​​are extracted, converted into matrix elements, and the elements are integrated into the overall graph data to generate a cross-institutional medical data association graph structure.

[0017] As a further solution of the present invention, based on the node and edge information in the cross-institutional medical data association graph structure, the connection strength of the nodes in the global graph is analyzed, the connection strength data is combined with the edge weight distribution parameter to filter the node weights, and the associated node information is eliminated according to the threshold range to generate the screening results of low-sensitivity nodes. Specifically, the steps are as follows:

[0018] S201: Based on the node and edge information in the cross-institutional medical data association graph structure, extract the connection data of each node, associate and integrate the number of node connections and the total weight, sort by the connection strength value range, and obtain node connection strength distribution data;

[0019] S202: Based on the node connection strength distribution data, call the node and edge weight data to match the edge weight distribution, analyze the node sensitivity score according to the node connection strength value, select the nodes whose sensitivity scores meet the threshold range, adjust the edge weight data distribution corresponding to the selected nodes, and generate a low-sensitivity node set;

[0020] S203: Based on the low-sensitivity node set, remove nodes and associated edges with sensitivity scores out of range, sort out the connection strength and edge weight information of the remaining nodes, integrate the filtered node and edge information into the graph structure, and generate a screening result of low-sensitivity nodes.

[0021] As a further solution of the present invention, based on the screening results of the low-sensitivity nodes, the perturbation amplitude analysis is performed according to the edge weight distribution parameters, and the node weight matrix is ​​reconstructed after adding the noise value in combination with the local subgraph weight ratio. The steps of generating the perturbed low-sensitivity node weight matrix are specifically as follows:

[0022] S301: Based on the screening results of the low-sensitivity nodes, extract the node weight value, analyze the distribution relationship between the node weight and the associated edge weight, calculate the cumulative value of the edge weight, analyze the impact of the edge weight parameter on the node weight, calculate the correlation value between the node weight and the edge weight, classify and integrate the weight and edge weight parameters, and obtain the node weight and edge weight relationship data;

[0023] S302: Based on the node weight and edge weight relationship data, extract the node weight ratio in the local subgraph, add a disturbance value according to the ratio, adjust the proportional relationship between the weight value after disturbance and the edge weight parameter, analyze the node weight value distribution, and generate a disturbance node weight matrix;

[0024] S303: Based on the perturbation node weight matrix, the perturbation weight value is mapped to the new matrix index according to the node global position, the matrix node arrangement and edge weight distribution structure are adjusted, and the perturbation weight data is sorted in combination with the global matrix to generate a perturbed low-sensitivity node weight matrix.

[0025] As a further solution of the present invention, the association value between the node weight and the edge weight is calculated according to the formula:

[0026]

[0027] Among them, W i Represents the association value between node weight and edge weight, P i represents the initial weight value of node i, E i Represents the total edge weight accumulated value of the edges associated with node i, C i represents the absolute value of the weight difference between node i and its directly connected nodes, Q i represents the average value of the global edge weight distribution, α is the edge weight enhancement adjustment coefficient, β is the edge weight smoothing adjustment coefficient, and γ is the weight difference mitigation coefficient.

[0028] As a further solution of the present invention, the node feature data in the matrix is ​​extracted by using the perturbed low-sensitivity node weight matrix, the correlation degree of the node embedding values ​​is analyzed by similarity comparison, the correlation value and the edge weight are recombined into weighted features, and the node-edge relationship is dynamically adjusted in combination with the weight offset to generate a feature-fused node-edge relationship matrix. Specifically, the following steps are performed:

[0029] S401: Utilizing the perturbed low-sensitivity node weight matrix, extracting node embedding values ​​and feature data, comparing node embedding values, analyzing the degree of association between adjacent nodes, combining the association values ​​with edge weight data, assigning weights according to the node embedding values, calculating the node embedding association weighted values, and obtaining the node embedding association weighted data;

[0030] S402: Based on the node embedding association weighted data, extract the weighted relationship between the node weight value and the edge weight data, adjust the node weight value according to the local subgraph node feature distribution, optimize the node feature value by combining the node weight data and the edge weight distribution, and generate a dynamically adjusted feature data matrix;

[0031] S403: Based on the dynamically adjusted feature data matrix, extract the optimized node weights and edge weight data, map the node positions according to the matrix global index, adjust the node feature data and edge weight distribution, and generate a feature-fused node-edge relationship matrix. As a further embodiment of the present invention, calculate the node embedding association weight value according to the formula:

[0032]

[0033] Among them, E w represents the node embedding association weight, Q j represents the weight of the edge between nodes, emb(u) and emb(v) represent the embedding vectors of nodes u and v respectively, and λ u and λ v Represents the eigenvalues ​​of nodes u and v, ∈ is a small positive number.

[0034] As a further solution of the present invention, based on the node-edge relationship matrix of the feature fusion, the steps of extracting the node feature distribution values ​​and edge weight distribution in the matrix, iteratively adjusting the weight parameters, statistically distributing and integrating the weight data, and generating the medical data privacy-preserving joint analysis distribution results are specifically as follows:

[0035] S501: Based on the node-edge relationship matrix of the feature fusion, extract node feature distribution values ​​and edge weights, analyze feature value distribution positions, reallocate weights and perform statistical classification to obtain node and edge feature distribution relationship data;

[0036] S502: Based on the node and edge feature distribution relationship data, analyze the node feature distribution deviation range, adjust the edge weight parameters multiple times according to the deviation range, analyze the node feature value distribution ratio, integrate the global node feature value and edge weight data, and obtain the global node weight and edge weight distribution data;

[0037] S503: Based on the global node weight and edge weight distribution data, extract the node and edge distribution parameters in the global matrix, adjust the mapping index between the node eigenvalue and the edge weight data, arrange the node and edge relationship data in the matrix, integrate the sorted distribution data, and generate the medical data privacy protection joint analysis distribution results.

[0038] On the other hand, an electric vehicle state monitoring system is provided, wherein the electric vehicle state monitoring system is used to execute the above electric vehicle state monitoring method, and the system includes:

[0039] The cross-institutional node association calculation module analyzes the correlation strength of sample attribute features based on the local medical data of medical institutions, compares the degree of feature matching to obtain a correlation value matrix, classifies and maps the matrix correlation values, and generates a cross-institutional node graph structure;

[0040] The sensitive node identification and elimination module extracts the node connection strength value and edge weight in the cross-institutional node graph structure, eliminates the nodes that exceed the range by comparing the sensitivity score with the set threshold range, and generates a low-sensitivity node feature set;

[0041] The weight perturbation and redistribution module extracts the node weight values ​​in the low-sensitivity node feature set, calculates the perturbation value by comparing the node weight ratio, and superimposes the perturbation value on the original weight parameter to redistribute the position in the weight matrix, and performs position mapping to generate a weight distribution adjustment matrix;

[0042] The embedding feature construction module adjusts the matrix according to the weight distribution, analyzes the weight and edge connection data, cross-combines the embedding feature values ​​and weight parameters, and performs mapping processing to adjust the connection features of nodes and edges to generate a node-edge embedding feature matrix;

[0043] The global feature distribution optimization module extracts the node feature distribution value and edge weight distribution in the node edge embedding feature matrix, adjusts the edge weight parameters by comparing the node feature value and the global feature deviation, and performs statistical integration to generate the medical data privacy protection joint analysis distribution result.

[0044] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0045] Through multi-dimensional analysis and processing of medical data nodes and their relationships, the efficiency of data privacy protection and joint analysis is optimized. By analyzing the correlations between sample features and constructing a cross-institutional correlation graph, the implicit correlations between data samples from different medical institutions can be accurately captured, enhancing the global perspective of data processing. Through matrix-based correlation processing, complex node correlations are converted into efficient mathematical expressions, effectively reducing redundant information interference during data analysis. By screening and reconstructing node weights, highly sensitive data nodes are eliminated, and the weight information of less sensitive nodes is perturbed. This combination of sensitivity scores and noise perturbation significantly improves privacy protection while preserving the analytical value of the data. The weight matrix of less sensitive nodes is dynamically adjusted to form new node-edge relationships. This feature fusion method further optimizes the accuracy of data correlations, enabling stronger feature matching during data analysis. By iteratively adjusting the global weight distribution and combining it with the dynamic relationship reconstruction between nodes, a more comprehensive statistical distribution integration is achieved, improving joint analysis capabilities under privacy protection conditions. This solution based on multi-level perturbation and dynamic adjustment significantly improves the strength and reliability of data privacy protection, while achieving better performance in cross-institutional collaboration in terms of data analysis accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0047] Figure 2 This is a detailed flow chart of S1 of the present invention;

[0048] Figure 3 This is a detailed flow chart of S2 of the present invention;

[0049] Figure 4 This is a detailed flow chart of S3 of the present invention;

[0050] Figure 5 This is a detailed flow chart of S4 of the present invention;

[0051] Figure 6 This is a detailed flow chart of S5 of the present invention;

[0052] Figure 7 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0053] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0054] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0055] In the embodiments of the present invention, the terms "image" and "picture" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same. The terms "of," "corresponding," and "corresponding" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same.

[0056] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0057] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0058] See also Figure 1 , an embodiment of the present invention provides a medical data analysis method based on blockchain and privacy protection, the processing flow of the method may include the following steps:

[0059] S1: Based on the local medical data of medical institutions, the correlation between sample features is analyzed to analyze the correlation between nodes. The data samples are used as nodes of the graph. The global homogeneity between subgraphs is cross-analyzed and the global correlation is converted into matrix elements to generate a cross-institutional medical data correlation graph structure.

[0060] S2: Based on the node and edge information in the cross-institutional medical data association graph structure, the connection strength of the nodes in the global graph is analyzed. The connection strength data is combined with the edge weight distribution parameters. The node weights are filtered according to the threshold range based on the sensitivity score, and the associated node information is eliminated to generate the screening results of low-sensitivity nodes.

[0061] S3: Based on the screening results of low-sensitivity nodes, extract the filtered node weight information, perform perturbation amplitude analysis on the node weight data according to the edge weight distribution parameters, combine the node weight proportion parameters in the local subgraph, add noise values ​​and reconstruct the node weight matrix, map the perturbed weights to the new matrix positions, and generate the perturbed low-sensitivity node weight matrix;

[0062] S4: Utilize the perturbed low-sensitivity node weight matrix to extract node feature data from the matrix. Analyze the correlation degree of node embedding values ​​through homogeneity comparison. Recombine the correlation value and edge weight into weighted features. Dynamically adjust the node-edge relationship by combining weight offset. Remap the adjusted feature data to generate a feature-fused node-edge relationship matrix.

[0063] S5: Based on the feature-fused node-edge relationship matrix, the node feature distribution values ​​and edge weight distribution in the matrix are extracted. The weight parameters are iteratively adjusted according to the feature distribution deviation. Combined with the global node relationship, the global weight data is statistically integrated to generate the medical data privacy-preserving joint analysis distribution results.

[0064] The cross-institutional medical data association graph structure includes node information, edge information, and a global homogeneity matrix. The screening results of low-sensitive nodes include the node weights after screening, the edge weight distribution parameters within the screening range, and the sensitivity score range. The perturbed low-sensitive node weight matrix includes the perturbed node weight values, node weight ratio parameters, and the perturbed position mapping matrix. The feature-fused node-edge relationship matrix includes the node feature distribution values, edge weight distribution, and dynamically adjusted weight features. The distribution results of the joint analysis of medical data privacy protection include the adjusted weight parameter distribution, the global node relationship statistical distribution, and the global weight integration results.

[0065] Specifically, if Figure 2 As shown in the figure, based on the local medical data of medical institutions, the correlation between nodes is analyzed through the correlation between sample features, the data samples are used as nodes of the graph, and the global homogeneity of the subgraphs is analyzed for similarity. The specific steps for generating the cross-institutional medical data association graph structure are as follows:

[0066] S101: Based on the local medical data of the medical institution, the data samples are parsed into independent nodes, the statistics of the numerical features are extracted, the value distribution of the categorical features is extracted, the feature interaction parameters are analyzed, the node association strength is filled into the matrix, and the sample association feature matrix is ​​generated;

[0067] Multiple feature dimensions are extracted independently from each sample data, and statistics such as the mean, median, standard deviation and skewness are calculated based on the numerical distribution of each dimension. The statistics are used as parameters to describe the numerical features. For categorical features, the frequency ratio of each value in the sample set is counted to obtain complete value distribution information. In order to analyze the interaction parameters between features, each two features are grouped together, and the amount of interaction information between each two features is calculated. The probability value of each feature combination is obtained by joint distribution, which reflects the degree of association between different features. A node association matrix is ​​constructed according to the interaction strength between features, in which the matrix elements represent the association strength of a specific feature combination. The association weights between samples are formed by accumulating the interaction information between each feature. When filling in the matrix elements, the intensity values ​​are denoised to eliminate the influence of outliers to ensure the accuracy of the matrix, and a sample association feature matrix is ​​generated as the input basis for feature interaction and node association in subsequent steps.

[0068] S102: Based on the sample association feature matrix, the node association strength values ​​in the matrix are converted according to the normalized range, the node adjacency relationships are extracted, the nodes are grouped according to the node adjacency parameters, the nodes in the group are assigned to subgraphs, the subgraph node layout is adjusted, and the sample subgraph structure model is generated;

[0069] Each association strength value in the matrix is ​​mapped to a specified range interval according to the ratio of the global maximum and minimum values, thereby ensuring the magnitude consistency of the association strength values ​​of all nodes. After normalization, the node adjacency relationship in the matrix is ​​further extracted. By calculating the adjacency parameters between nodes, node pairs with association strength higher than a certain threshold are marked as directly adjacent. At the same time, the adjacency is graded and labeled according to the association strength value. Based on the adjacency parameter of each node, a node grouping rule is constructed, and nodes with higher adjacency parameters are assigned to the same group. The strength values ​​between nodes in each group are reordered to generate a group node list. By constructing subgraphs and adjusting the node layout, the nodes in the group are rearranged so that their layout can optimally reflect the node association strength and grouping relationship, and the integrity and connectivity of the local structure are retained in each subgraph. Finally, a sample subgraph structure model is generated, providing a data basis for global matching and integration between subgraphs.

[0070] S103: Based on the sample subgraph structure model, analyze the node matching relationship between subgraphs, calculate the cross-subgraph node connection parameters and association strength values, extract the global subgraph homogeneity values, convert them into matrix elements, integrate the elements into the overall graph data, and generate the cross-institutional medical data association graph structure;

[0071] By comparing the eigenvalues ​​of shared nodes between subgraphs, the node matching degree across subgraphs is gradually calculated. The matching degree value is determined by the homogeneity of the feature interaction weights, and the weight is adjusted by the strength value of the association matrix calculated previously. The node connection parameters across subgraphs are calculated, and the association strength value of each pair of matching nodes is evaluated one by one. The strength value is corrected based on their positional relationship across the subgraph to ensure the connectivity after matching across subgraphs. When extracting the global homogeneity value of the subgraph, the matching node ratio and the corresponding association strength value between the subgraphs are weighted and integrated into a global homogeneity value, which is recorded in the corresponding elements of the matrix to form a global homogeneity matrix. By integrating all elements in the matrix into the overall graph data, different subgraphs are connected into an overall association graph structure, thereby generating a cross-institutional medical data association graph structure, providing support for data privacy protection and distributed analysis based on federated learning, while ensuring the security and integrity of medical data across institutions.

[0072] Specifically, if Figure 3 As shown in the figure, based on the node and edge information in the cross-institutional medical data association graph structure, the connection strength of the nodes in the global graph is analyzed, the connection strength data is combined with the edge weight distribution parameters to filter the node weights, and the associated node information is eliminated according to the threshold range. The specific steps for generating the screening results of low-sensitivity nodes are as follows:

[0073] S201: Based on the node and edge information in the cross-institutional medical data association graph structure, extract the connection data of each node, associate and integrate the number of node connections and the total weight, sort by the connection strength value range, and obtain the node connection strength distribution data;

[0074] Taking each node in the association graph as a unit, the number of all directly adjacent nodes of the node is counted, and the sum of the corresponding edge weights in the adjacency relationship is calculated, where the edge weight is obtained through the previously generated association matrix elements. In order to more intuitively reflect the importance of the node, the number of connections of each node and the sum of its edge weights are weighted and integrated to generate the connection strength value of the node, and the nodes are sorted according to the size of the strength value to identify the nodes with greater influence in the network. The connection strength distribution is segmented and counted, and the nodes are divided into several range intervals according to the strength value distribution. The number of nodes and the corresponding sum of edge weights are recorded in each interval to obtain the node connection strength distribution data. The distribution data is used to describe the importance and relevance of nodes in the network, providing data support for subsequent node screening and sensitivity analysis.

[0075] S202: Based on the node connection strength distribution data, call the node and edge weight data to match the edge weight distribution, analyze the node sensitivity score according to the node connection strength value, select the nodes whose sensitivity scores meet the threshold range, adjust the edge weight data distribution corresponding to the selected nodes, and generate a low-sensitivity node set;

[0076] According to the connection strength distribution range of each node, the corresponding edge weight is analyzed. By calculating the weighted average of the adjacent edge weights of the node, the edge weight distribution index of each node is generated. On this basis, the sensitivity score of the node is analyzed according to the node connection strength value range. The sensitivity score uses the node connection strength and edge weight distribution as input parameters. The comprehensive weight reflects the privacy sensitivity of the node in the overall network. By setting the sensitivity threshold range, the nodes whose sensitivity scores meet the threshold range are screened. At the same time, the edge weight data distribution corresponding to the screened nodes is readjusted, and the edge weights that are too high or too low are normalized and mapped to reduce the impact of local abnormal data. Finally, a low-sensitivity node set is generated. This node set contains all nodes whose sensitivity scores meet the threshold range and their associated edge information, providing input data for subsequent privacy protection processing and graph structure adjustment.

[0077] S203: Based on the low-sensitivity node set, remove nodes and associated edges with sensitivity scores outside the range, sort out the connection strength and edge weight information of the remaining nodes, integrate the filtered node and edge information into the graph structure, and generate a low-sensitivity node screening result;

[0078] The sensitivity scores of all nodes are compared one by one, and nodes that exceed the threshold range are screened out and removed from the network structure. At the same time, all edge information directly associated with the nodes is removed, and the data of the remaining nodes are sorted out. The node connection strength and edge weight information are recalculated to ensure that the connection strength values ​​and edge weight sums of the remaining nodes remain complete and consistent. In the screened node set and edge information, the graph structure is readjusted according to the connection strength distribution and edge weight distribution data of the remaining nodes, so that the low-sensitivity node set forms a new topological layout in the network structure. By integrating the screened nodes and edge information, the low-sensitivity node screening results are generated. This result provides efficient and secure basic data support for further privacy protection and data analysis.

[0079] Specifically, if Figure 4 As shown in the figure, based on the screening results of low-sensitivity nodes, the perturbation amplitude analysis is performed according to the edge weight distribution parameters. The node weight matrix is ​​reconstructed after adding the noise value based on the local subgraph weight ratio. The specific steps for generating the perturbed low-sensitivity node weight matrix are as follows:

[0080] S301: Based on the screening results of low-sensitivity nodes, extract node weight values, analyze the distribution relationship between node weights and associated edge weights, calculate the cumulative value of edge weights, analyze the impact of edge weight parameters on node weights, calculate the correlation value between node weights and edge weights, classify and integrate weight and edge weight parameters, and obtain node weight and edge weight relationship data;

[0081] Calculate the association value between node weight and edge weight according to the formula:

[0082]

[0083] Among them, W i Represents the association value between node weight and edge weight, P i represents the initial weight value of node i, E i Represents the total edge weight accumulated value of the edges associated with node i, C i represents the absolute value of the weight difference between node i and its directly connected nodes, Q i represents the average value of the global edge weight distribution, α is the edge weight enhancement adjustment coefficient, β is the edge weight smoothing adjustment coefficient, and γ is the weight difference mitigation coefficient;

[0084] This formula is used to calculate the comprehensive weight value W of node i i ,The results are used to evaluate the global importance of nodes in low-sensitivity environments,

[0085] Parameter meaning and setting value:

[0086] P i : The initial weight value of node i, obtained by normalizing the node features, is set to 0.6, reflecting the importance of the node in the initial state;

[0087] E i : The total edge weight accumulated value of the edges associated with node i, obtained by summing the weights of all edges directly connected to node i, is set to 1.2, reflecting the closeness of the connection between nodes;

[0088] C i : The absolute value of the weight difference between node i and its directly connected nodes. It is obtained by calculating the absolute value of the difference between the weight of node i and the weight of its directly connected nodes and averaging all connected nodes. It is set to 0.3, reflecting the difference in importance between the node and its neighboring nodes.

[0089] Q i : The average value of the global edge weight distribution, which is obtained by calculating the average weight of all edges in the network and is set to 0.5, reflecting the overall level of edge weight in the network;

[0090] α: Edge weight enhancement adjustment coefficient, used to control the strength of the association between nodes and edge weight distribution, set to 1.5. This value fluctuates with the importance of edge weights in the network;

[0091] β: Edge weight smoothing adjustment coefficient, used to reduce the fluctuation of calculation results caused by large differences in edge weights. It is set to 0.4. This value fluctuates with the degree of difference in edge weights in the network.

[0092] γ: Weight difference mitigation coefficient, used to smooth the impact of weight differences between nodes on the final result. It is set to 0.7 and fluctuates with the degree of node weight difference.

[0093] Substitute the parameters into the formula for calculation:

[0094] Calculate|P i ·E i |:|P i ·E i |=|0.6×1.2|=0.72;

[0095] calculate

[0096] Calculate|C i -Q i |:|C i -Q i |=|0.3-0.5|=0.2;

[0097] calculate

[0098] Calculate the denominator:

[0099] Calculate the molecular part:

[0100] Finally calculate W i :

[0101] Results i ≈1.856 indicates that the comprehensive weight value of node i is 1.856, reflecting the global importance of node i in a low-sensitivity environment, taking into account the initial weight of the node, the weight of the associated edge, the weight difference between nodes, and the various adjustment coefficients.

[0102] S302: Based on the node weight and edge weight relationship data, extract the node weight ratio in the local subgraph, add a perturbation value according to the ratio, adjust the proportional relationship between the perturbed weight value and the edge weight parameter, analyze the node weight value distribution, and generate a perturbed node weight matrix;

[0103] In each local subgraph, the ratio of the node weight value to the total weight value of the local subgraph is calculated and stored one by one to form a node weight ratio sequence. A disturbance value is added to each weight ratio. By setting a certain range interval for the disturbance value, it is ensured that the numerical change of the weight after disturbance will not exceed the sensitivity threshold range. At the same time, the influence of the weight disturbance of each node on the edge weight parameter is controlled within an acceptable range. After adding the disturbance value, the proportional relationship between the adjusted node weight value and the edge weight parameter is recalculated, and the adjusted data is analyzed. By further analyzing the distribution of all node weights, a disturbance node weight matrix is ​​established. The elements in the matrix record the adjustment results of the node weight and its associated edge weight after disturbance, providing support for subsequent processing.

[0104] S303: Based on the perturbation node weight matrix, the perturbation weight value is mapped to the new matrix index according to the node global position, the matrix node arrangement and edge weight distribution structure are adjusted, and the perturbation weight data is sorted in combination with the global matrix to generate the perturbed low-sensitivity node weight matrix;

[0105] According to the position sequence of the node in the global graph structure, the perturbed node weight value is remapped to the corresponding index position of the new matrix one by one to maintain the integrity of the global matrix structure. After completing the node weight mapping, the node arrangement and edge weight distribution structure in the matrix are adjusted. By rearranging the matrix node index order and the distribution of the associated edge weights, the global consistency of the nodes and edge weights is ensured. The perturbed node weight data is combined with the original global matrix data. By comparing and merging the two matrices, a perturbed low-sensitivity node weight matrix is ​​generated. This matrix contains the perturbed node weight values ​​and the adjusted global association information, providing complete basic data input for privacy protection and distributed modeling of medical data in federated learning.

[0106] Specifically, if Figure 5 As shown in the figure, the perturbed low-sensitivity node weight matrix is ​​used to extract node feature data from the matrix. The correlation degree of node embedding values ​​is analyzed through homogeneity comparison. The correlation value and edge weight are recombined into weighted features. The node and edge relationships are dynamically adjusted in combination with weight offset. The specific steps for generating the feature-fused node-edge relationship matrix are as follows:

[0107] S401: Using the perturbed low-sensitivity node weight matrix, extract node embedding values ​​and feature data, compare node embedding values, analyze the degree of association between adjacent nodes, combine the association values ​​with edge weight data, assign weights according to node embedding values, calculate node embedding association weighted values, and obtain node embedding association weighted data;

[0108] Calculate the node embedding association weight value according to the formula:

[0109]

[0110] Among them, E w represents the node embedding association weight, Q j represents the weight of the edge between nodes, emb(u) and emb(v) represent the embedding vectors of nodes u and v respectively, and λ u and λ v Represents the eigenvalues ​​of nodes u and v, ∈ is a small positive number;

[0111] This formula is used to calculate the node embedding association weighted data, and the result is used to obtain the weighted association matrix;

[0112] Q jIt represents the weight of the edge between nodes, reflecting the direct connection strength between nodes. It is measured by the interaction frequency between nodes in the actual network. For example, if the interaction frequency between nodes is 10 times per hour, then Q j =10;

[0113] emb(u) and emb(v) represent the embedding vectors of nodes u and v, respectively. They are used to capture the structure and feature information of the nodes and are calculated using a graph embedding algorithm (e.g., Node2Vec). For example, the embedding vector of node u is emb(u) = 0.8, and the embedding vector of node v is emb(v) = 0.6.

[0114] λ u and λ v Represents the eigenvalues ​​of nodes u and v, reflecting the structural sensitivity of the nodes in the graph. It is obtained by calculating the degree centrality of the nodes. If the degree of node u is 5 and the degree of node v is 3, then λ u =5,λ v =3;

[0115] ∈ is a small positive number used to avoid the denominator being zero and increase the stability of the calculation. It is set to a minimum value, such as ∈=10 -6 , to ensure computational stability;

[0116] Substitute the parameters into the formula for calculation:

[0117] Compute the sum of the products of the weights and the embedding vector:

[0118] Since only one side is considered here, let w1 = 10, then: 10 × 0.8 × 0.6 = 4.8;

[0119] Calculate the absolute value of the eigenvalue difference and add ∈:|λ u -λ v |+∈=|5-3|+10 -6 =2+10 -6 ;

[0120] Calculate the square root of the denominator:

[0121] Calculate the final weighted correlation value E w :

[0122] Results E w ≈3.394 indicates that the embedding association weighted value between node u and node v is 3.394, which is used to construct the weighted association matrix to reflect the association strength between nodes.

[0123] S402: Based on the node embedding associated weighted data, extract the weighted relationship between the node weight value and the edge weight data, adjust the node weight value according to the local subgraph node feature distribution, optimize the node feature value by combining the node weight data and the edge weight distribution, and generate a dynamically adjusted feature data matrix;

[0124] For each node, the weighted average of its weight value and the weight of the adjacent edges is calculated to generate the comprehensive weighted relationship parameters of the nodes and edges. Within the local subgraph, the node feature distribution is analyzed, and the node weight value is adjusted according to the node's embedded position and feature distribution in the subgraph, so that the adjusted weight value can better adapt to the local subgraph feature changes. The node weight adjustment result is integrated with the edge weight distribution data. By recalculating the weight ratio of the node and the edge, the node's eigenvalue is optimized to maintain high consistency and effectiveness both in the subgraph and globally. After the optimization is completed, the eigenvalues ​​of all nodes are dynamically adjusted and stored in the feature data matrix, where each matrix element records the optimized eigenvalue of the node and its associated edge weight data, laying the foundation for global feature fusion and matrix adjustment.

[0125] S403: Based on the dynamically adjusted feature data matrix, extract the optimized node weights and edge weight data, map the node positions according to the matrix global index, adjust the node feature data and edge weight distribution, and generate a feature-fused node-edge relationship matrix;

[0126] By extracting the optimized weight values ​​and edge weights of each node in the matrix one by one, and combining the global index information of the matrix, the node positions are globally mapped to ensure that the optimized node weight and edge weight distribution can be consistent with the global graph structure. After the node positions are mapped, the node feature data and edge weight distribution are adjusted as a whole. By rearranging the index order of nodes and edges, the optimized weight values ​​and feature distribution can fully reflect the global characteristics of the graph structure. After the adjustment is completed, the optimized node feature values ​​and edge weight data are integrated into a unified feature fusion matrix, in which each matrix element contains the comprehensive feature relationship between nodes and edges, and the rows and columns of the matrix correspond to the global indexes of nodes and edges respectively, generating a feature-fused node-edge relationship matrix, which provides the final optimized data support for medical data privacy protection and analysis based on federated learning, ensuring the security and effectiveness of data use.

[0127] Specifically, if Figure 6 As shown in the figure, based on the node-edge relationship matrix of feature fusion, the steps of extracting the node feature distribution values ​​and edge weight distribution in the matrix, iteratively adjusting the weight parameters, statistically distributing and integrating the weight data, and generating the distribution results of the medical data privacy-preserving joint analysis are as follows:

[0128] S501: Based on the node-edge relationship matrix of feature fusion, extract the node feature distribution value and edge weight, analyze the feature value distribution position, redistribute the weight and perform statistical classification to obtain the node and edge feature distribution relationship data;

[0129] The characteristic distribution value of each node in the matrix is ​​read one by one, and the eigenvalue is mapped and analyzed according to its corresponding edge weight, and the weight relationship between the node eigenvalue and its adjacent edge is integrated into distribution data. When analyzing the eigenvalue distribution position, the position of its eigenvalue in the overall distribution is analyzed node by node according to the row and column index of the matrix. The distribution level of the node eigenvalue is determined by comparing the node eigenvalue with the global mean of the characteristic distribution. After completing the distribution position analysis, the weights of the nodes and edges are redistributed. By combining the relative relationship between the node eigenvalue and the edge weight, the weight adjustment coefficient of each node is generated, and the node weight value is corrected according to the adjustment coefficient. At the same time, the proportional relationship between the node eigenvalue and the edge weight is redistributed. The characteristic distribution and edge weight data of all nodes are counted, and the data is classified according to the distribution level to form the node and edge characteristic distribution relationship data, which provides input basis for subsequent distribution deviation analysis and global data integration.

[0130] S502: Based on the node and edge feature distribution relationship data, analyze the node feature distribution deviation range, adjust the edge weight parameters multiple times according to the deviation range, analyze the node feature value distribution ratio, integrate the global node feature value and edge weight data, and obtain the global node weight and edge weight distribution data;

[0131] The difference between the distribution data of each node's eigenvalue and the mean of the global feature distribution is calculated to obtain the deviation value of the node feature distribution. The larger the deviation value, the higher the degree of deviation between the node feature distribution and the global distribution. Nodes with high deviation ranges are classified according to the deviation range, and multiple adjustments are made to the nodes. The deviation value is gradually reduced by progressively correcting the adjacent edge weight parameters. After each edge weight parameter adjustment, the distribution ratio of the node eigenvalue is recalculated, and the ratio change is integrated into the node eigenvalue update result. For nodes in the low deviation range, the node eigenvalue and edge weight data are fine-tuned by combining the global feature distribution ratio to ensure the uniformity of the overall feature distribution. After the adjustment is completed, the global node eigenvalue and edge weight data are integrated into a unified distribution data structure to generate global node weight and edge weight distribution data, laying the data foundation for subsequent distribution parameter mapping and matrix organization.

[0132] S503: Based on the global node weight and edge weight distribution data, extract the node and edge distribution parameters in the global matrix, adjust the mapping index between the node eigenvalue and the edge weight data, arrange the node and edge relationship data in the matrix, integrate the sorted distribution data, and generate the medical data privacy-preserving joint analysis distribution results;

[0133] The weight value and its corresponding edge weight parameter are extracted for each node in the matrix one by one, and the node index in the matrix is ​​readjusted according to the distribution parameter. By rearranging the node index position, the node and edge distribution in the matrix is ​​optimized. When adjusting the mapping index of the node eigenvalue and the edge weight data, the eigenvalue and edge weight parameter are remapped to the global coordinate position corresponding to the matrix index based on the global distribution characteristics of the node, ensuring that the adjusted distribution can reflect the actual relationship between the node and the edge. After completing the adjustment of the node and edge relationship data, the sorted distribution data in the global matrix is ​​classified and integrated to generate the medical data privacy protection joint analysis distribution result. This distribution result not only reflects the global correlation between nodes and edges, but also provides an efficient and secure data distribution method for privacy protection under the federated learning framework.

[0134] like Figure 7 As shown, a medical data analysis system based on blockchain and privacy protection includes:

[0135] The cross-institutional node association calculation module analyzes the correlation strength of sample attribute features based on the local medical data of medical institutions, compares the degree of feature matching to obtain a correlation value matrix, classifies and maps the matrix correlation values, and generates a cross-institutional node graph structure;

[0136] The sensitive node identification and elimination module extracts the node connection strength value and edge weight in the cross-institutional node graph structure, and eliminates the nodes that exceed the range by comparing the sensitivity score with the set threshold range to generate a low-sensitivity node feature set;

[0137] The weight perturbation and redistribution module extracts the node weight values ​​from the low-sensitivity node feature set, calculates the perturbation value by comparing the node weight ratio, and superimposes the perturbation value on the original weight parameter to redistribute the position in the weight matrix and perform position mapping to generate a weight distribution adjustment matrix.

[0138] The embedding feature construction module adjusts the matrix according to the weight distribution, analyzes the weight and edge connection data, cross-combines the embedding feature values ​​and weight parameters, and performs mapping processing to adjust the connection characteristics of nodes and edges to generate a node-edge embedding feature matrix;

[0139] The global feature distribution optimization module extracts the node feature distribution values ​​and edge weight distributions from the node-edge embedding feature matrix, adjusts the edge weight parameters by comparing the node feature values ​​and the global feature deviation, and performs statistical integration to generate the medical data privacy-preserving joint analysis distribution results.

[0140] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A medical data analysis method based on blockchain and privacy protection, characterized in that: The following steps are involved: S1: Based on the local medical data of medical institutions, the correlation between nodes is analyzed through the correlation between sample features. The data samples are used as nodes of the graph. The global homogeneity of the subgraphs is analyzed in the same way to generate the cross-institutional medical data association graph structure. S2: Analyze the connection strength of the nodes in the global graph based on the node and edge information in the cross-institutional medical data association graph structure, combine the connection strength data with the edge weight distribution parameters to filter the node weights, eliminate the associated node information according to the threshold range, and generate a screening result of low-sensitivity nodes; S3: Based on the screening results of the low-sensitivity nodes, a disturbance amplitude analysis is performed according to the edge weight distribution parameters, and the node weight matrix is ​​reconstructed after adding the noise value based on the local subgraph weight ratio to generate a perturbed low-sensitivity node weight matrix; S4: Using the perturbed low-sensitivity node weight matrix, extracting node feature data in the matrix, analyzing the correlation degree of node embedding values ​​through homogeneity comparison, recombining the correlation value and edge weight into weighted features, dynamically adjusting the node and edge relationship in combination with weight offset, and generating a feature-fused node-edge relationship matrix; S5: Based on the node-edge relationship matrix of the feature fusion, extract the node feature distribution values ​​and edge weight distribution in the matrix, iteratively adjust the weight parameters, statistically distribute and integrate the weight data, and generate the medical data privacy protection joint analysis distribution results.

2. A medical data analysis method based on blockchain and privacy protection according to claim 1, characterized in that: The cross-institutional medical data association graph structure includes node information, edge information, and a global homogeneity matrix. The screening results of the low-sensitive nodes include the node weights after screening, the edge weight distribution parameters within the screening range, and the sensitivity score range. The perturbed low-sensitive node weight matrix includes the perturbed node weight values, the node weight ratio parameters, and the perturbed position mapping matrix. The feature-fused node-edge relationship matrix includes the node feature distribution values, the edge weight distribution, and the dynamically adjusted weight features. The medical data privacy protection joint analysis distribution results include the adjusted weight parameter distribution, the global node relationship statistical distribution, and the global weight integration results.

3. A medical data analysis method based on blockchain and privacy protection according to claim 1, characterized in that: Based on the local medical data of medical institutions, the correlation between nodes is analyzed through the association between sample features. The data samples are used as nodes of the graph. The global homogeneity of the subgraphs is analyzed by similarity. The specific steps to generate the cross-institutional medical data association graph structure are as follows: S101: Based on the local medical data of the medical institution, the data samples are parsed into independent nodes, the statistics of the numerical features are extracted, the value distribution of the categorical features is extracted, the feature interaction parameters are analyzed, the node association strength is filled into the matrix, and the sample association feature matrix is ​​generated; S102: Based on the sample association feature matrix, convert the node association strength values ​​in the matrix according to a normalized range, extract the node adjacency relationship, group the nodes according to the node adjacency parameters, assign the nodes in the group to subgraphs, adjust the subgraph node layout, and generate a sample subgraph structure model; S103: Based on the sample subgraph structure model, the node matching relationship between subgraphs is analyzed, the cross-subgraph node connection parameters and association strength values ​​are calculated, the global subgraph homogeneity values ​​are extracted, converted into matrix elements, and the elements are integrated into the overall graph data to generate a cross-institutional medical data association graph structure.

4. A medical data analysis method based on blockchain and privacy protection according to claim 1, characterized in that: The steps of analyzing the connection strength of the nodes in the global graph based on the node and edge information in the cross-institutional medical data association graph structure, combining the connection strength data with the edge weight distribution parameters to filter the node weights, and eliminating the associated node information according to the threshold range to generate the screening results of low-sensitivity nodes are as follows: S201: Based on the node and edge information in the cross-institutional medical data association graph structure, extract the connection data of each node, associate and integrate the number of node connections and the total weight, sort by the connection strength value range, and obtain node connection strength distribution data; S202: Based on the node connection strength distribution data, call the node and edge weight data to match the edge weight distribution, analyze the node sensitivity score according to the node connection strength value, select the nodes whose sensitivity scores meet the threshold range, adjust the edge weight data distribution corresponding to the selected nodes, and generate a low-sensitivity node set; S203: Based on the low-sensitivity node set, remove nodes and associated edges with sensitivity scores out of range, sort out the connection strength and edge weight information of the remaining nodes, integrate the filtered node and edge information into the graph structure, and generate a screening result of low-sensitivity nodes.

5. A medical data analysis method based on blockchain and privacy protection according to claim 1, characterized in that: Based on the screening results of the low-sensitivity nodes, the perturbation amplitude analysis is performed according to the edge weight distribution parameters. The node weight matrix is ​​reconstructed after adding the noise value based on the local subgraph weight ratio. The steps of generating the perturbed low-sensitivity node weight matrix are as follows: S301: Based on the screening results of the low-sensitivity nodes, extract the node weight value, analyze the distribution relationship between the node weight and the associated edge weight, calculate the cumulative value of the edge weight, analyze the impact of the edge weight parameter on the node weight, calculate the correlation value between the node weight and the edge weight, classify and integrate the weight and edge weight parameters, and obtain the node weight and edge weight relationship data; S302: Based on the node weight and edge weight relationship data, extract the node weight ratio in the local subgraph, add a disturbance value according to the ratio, adjust the proportional relationship between the weight value after disturbance and the edge weight parameter, analyze the node weight value distribution, and generate a disturbance node weight matrix; S303: Based on the perturbation node weight matrix, the perturbation weight value is mapped to the new matrix index according to the node global position, the matrix node arrangement and edge weight distribution structure are adjusted, and the perturbation weight data is sorted in combination with the global matrix to generate a perturbed low-sensitivity node weight matrix.

6. A medical data analysis method based on blockchain and privacy protection according to claim 5, characterized in that: Calculate the association value between the node weight and the edge weight according to the formula: Among them, W i Represents the association value between node weight and edge weight, P i represents the initial weight value of node i, E i Represents the total edge weight accumulated value of the edges associated with node i, C i represents the absolute value of the weight difference between node i and its directly connected nodes, Q i represents the average value of the global edge weight distribution, α is the edge weight enhancement adjustment coefficient, β is the edge weight smoothing adjustment coefficient, and γ is the weight difference mitigation coefficient.

7. A medical data analysis method based on blockchain and privacy protection according to claim 1, characterized in that: The perturbed low-sensitivity node weight matrix is ​​used to extract node feature data from the matrix. The correlation degree of node embedding values ​​is analyzed by similarity comparison. The correlation value and edge weight are recombined into weighted features. The node and edge relationships are dynamically adjusted in combination with weight offset to generate a feature-fused node-edge relationship matrix. The specific steps are as follows: S401: Utilizing the perturbed low-sensitivity node weight matrix, extracting node embedding values ​​and feature data, comparing node embedding values, analyzing the degree of association between adjacent nodes, combining the association values ​​with edge weight data, assigning weights according to the node embedding values, calculating the node embedding association weighted values, and obtaining the node embedding association weighted data; S402: Based on the node embedding association weighted data, extract the weighted relationship between the node weight value and the edge weight data, adjust the node weight value according to the local subgraph node feature distribution, optimize the node feature value by combining the node weight data and the edge weight distribution, and generate a dynamically adjusted feature data matrix; S403: Based on the dynamically adjusted feature data matrix, extract the optimized node weights and edge weight data, map the node positions according to the matrix global index, adjust the node feature data and edge weight distribution, and generate a feature-fused node-edge relationship matrix.

8. A medical data analysis method based on blockchain and privacy protection according to claim 7, characterized in that: Calculate the node embedding association weighted value according to the formula: Among them, E w represents the node embedding association weight, Q j represents the weight of the edge between nodes, emb(u) and emb(v) represent the embedding vectors of nodes u and v respectively, and λ u and λ v Represents the eigenvalues ​​of nodes u and v, ∈ is a small positive number.

9. A medical data analysis method based on blockchain and privacy protection according to claim 1, characterized in that: Based on the node-edge relationship matrix of the feature fusion, the node feature distribution values ​​and edge weight distribution in the matrix are extracted, the weight parameters are iteratively adjusted, and the weight data is statistically distributed and integrated to generate the distribution results of the medical data privacy-preserving joint analysis. The specific steps are: S501: Based on the node-edge relationship matrix of the feature fusion, extract node feature distribution values ​​and edge weights, analyze feature value distribution positions, reallocate weights and perform statistical classification to obtain node and edge feature distribution relationship data; S502: Based on the node and edge feature distribution relationship data, analyze the node feature distribution deviation range, adjust the edge weight parameters multiple times according to the deviation range, analyze the node feature value distribution ratio, integrate the global node feature value and edge weight data, and obtain the global node weight and edge weight distribution data; S503: Based on the global node weight and edge weight distribution data, extract the node and edge distribution parameters in the global matrix, adjust the mapping index between the node eigenvalue and the edge weight data, arrange the node and edge relationship data in the matrix, integrate the sorted distribution data, and generate the medical data privacy protection joint analysis distribution results.

10. A medical data analysis system based on blockchain and privacy protection, characterized in that: A method for analyzing medical data based on blockchain and privacy protection according to any one of claims 1 to 9, wherein the system comprises: The cross-institutional node association calculation module analyzes the correlation strength of sample attribute features based on the local medical data of medical institutions, compares the degree of feature matching to obtain a correlation value matrix, classifies and maps the matrix correlation values, and generates a cross-institutional node graph structure; The sensitive node identification and elimination module extracts the node connection strength value and edge weight in the cross-institutional node graph structure, eliminates the nodes that exceed the range by comparing the sensitivity score with the set threshold range, and generates a low-sensitivity node feature set; The weight perturbation and redistribution module extracts the node weight values ​​in the low-sensitivity node feature set, calculates the perturbation value by comparing the node weight ratio, and superimposes the perturbation value on the original weight parameter to redistribute the position in the weight matrix, and performs position mapping to generate a weight distribution adjustment matrix; The embedding feature construction module adjusts the matrix according to the weight distribution, analyzes the weight and edge connection data, cross-combines the embedding feature values ​​and weight parameters, and performs mapping processing to adjust the connection features of nodes and edges to generate a node-edge embedding feature matrix; The global feature distribution optimization module extracts the node feature distribution value and edge weight distribution in the node edge embedding feature matrix, adjusts the edge weight parameters by comparing the node feature value and the global feature deviation, and performs statistical integration to generate the medical data privacy protection joint analysis distribution result.

Images