Block chain communication method and system for bone tumor data security sharing

Abstract

The invention relates to a block chain communication method and system for bone tumor data security sharing, and particularly relates to the field of payment protocols and security transactions for business management, and the scheme realizes verification of the accuracy and foresight of node screening by constructing a dynamic evaluation system fusing node behavior prediction and mechanism situational trust, thereby improving the security of bone tumor data. The method greatly reduces verification delay, effectively avoids resource redundancy and waste based on an elastic verification scale mechanism of a trust state and a transaction risk, remarkably improves consensus efficiency, introduces real-time performance monitoring and double-loop feedback adjustment, enables the system to have a continuous self-optimization capability, can dynamically adapt to node behavior change and network fluctuation, and improves the reliability of the system. Therefore, efficient and reliable intelligent collaborative verification is achieved while high security of bone tumor data sharing transaction is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of payment protocols and secure transactions for business management, and more specifically, to a blockchain communication method and system for secure sharing of bone tumor data. Background Technology

[0002] With the development of precision medicine and interdisciplinary research, multi-center scientific research collaborations for complex diseases such as bone tumors are becoming increasingly frequent. Against this backdrop, diverse entities such as medical institutions, independent laboratories, and pharmaceutical R&D companies have formed close-knit medical research alliances. To maximize the value of scientific research, there is a strong demand for data sharing among the institutions within the alliance. For example, to advance the development of drugs targeting specific targets, a research institution may need to purchase limited-term access to a specific dataset from another medical institution that possesses valuable bone tumor imaging and genomic data. Such sharing behavior is essentially a small-amount, high-frequency transfer of data usage rights and value, involving complex ownership confirmation, payment settlement, and immediate access authorization. However, the heterogeneity of information systems within each institution, independent data management strategies, and the strict medical data compliance regulations such as GDPR and HIPAA pose serious challenges to the traditional sharing model based on centralized data platforms or simple point-to-point transmission in terms of security, reliability, and collaborative efficiency.

[0003] To address these challenges, existing technologies have included solutions for building medical data sharing platforms using blockchain technology. These solutions leverage the distributed ledger and smart contract capabilities of blockchain to achieve immutability and automated execution of data transaction records. However, existing technologies have significant shortcomings in the specific cross-institutional transaction verification process. Taking a typical blockchain-based medical data management system as an example, when processing a data access permission transaction, it typically needs to broadcast the transaction to all nodes in the blockchain network and rely on consensus algorithms such as Proof-of-Work or Proof-of-Stake for synchronous verification of the entire network state and block confirmation. While this process ensures decentralized security, it introduces significant latency. This approach struggles to meet the urgent need for real-time or near-real-time data acquisition in scientific research collaborations. Furthermore, to meet the privacy compliance requirements of medical data, transactions often require the integration of complex cross-domain identity authentication and zero-knowledge verification mechanisms, further increasing the complexity and time consumption of the verification process. While centralized payment gateway solutions offer advantages in transaction speed, they cannot simultaneously guarantee data sovereignty and payment security due to the introduction of trusted third parties, posing single-point failure and data leakage risks. Therefore, the current technological bottleneck lies in the lack of a dedicated protocol and system architecture that can be deeply integrated into bone tumor data sharing scenarios and achieve efficient and lightweight verification of cross-institutional transactions while ensuring data security and privacy compliance. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing a blockchain communication method and system for secure sharing of bone tumor data, thereby resolving the issues raised in the background section.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: a blockchain communication method for secure sharing of bone tumor data, specifically including the following steps: Step S1: In response to receiving a bone tumor data sharing transaction verification request in the blockchain network, based on the historical transaction verification records of the verification nodes, construct and update a machine learning model for quantifying the reputation of the verification nodes and predicting their verification behavior; the historical transaction verification records include the historical verification success rate, historical average response latency, and participation frequency distribution of each verification node in the verification of different subtypes of bone tumor data transactions; the machine learning model analyzes the historical transaction verification records and outputs the dynamic reputation score of each verification node and the expected behavior evaluation result for the current transaction; Step S2: Based on the dynamic reputation score of the verification node, the attribute information of the institution to which the verification node belongs, and the specific contextual features of the current transaction, a dynamic two-layer trust network is generated; the institution attribute information includes historical cooperation data between institutions; the physical node layer of the two-layer trust network corresponds to each verification node, and the logical institution layer corresponds to the institution to which each verification node belongs; using graph embedding technology and attention mechanism, combined with the specific contextual features of the current transaction, the contextual trust degree of each institution in this transaction is calculated. Step S3: Based on the expected behavior assessment results of the verification nodes and the contextual trust level of the institution, execute optimization decisions. Optimization decisions include: selecting verification nodes from the verification nodes to form a dynamic verification node set based on the contextual trust level of the institution and the expected behavior assessment results of the verification nodes, and assigning verification tasks to the verification nodes in the dynamic verification node set. The size of the dynamic verification node set is calculated using a dynamic node number determination formula based on the network average trust confidence level calculated from the contextual trust level and the transaction risk score calculated by extracting the transaction amount and data sensitivity level from the current transaction information. The dynamic node number determination formula includes an adjustable size adjustment coefficient. Step S4: During the verification process of the dynamic verification node set, collect actual verification performance data in real time. The actual verification performance data includes consensus achievement time, node response consistency, and node response latency. Compare the actual verification performance data with the expected behavior evaluation results output by the machine learning model in step S1. Based on the differences generated by the comparison, execute a feedback adjustment process. The feedback adjustment process is used to adjust the machine learning model in step S1 and to adjust the scale adjustment coefficient included in the dynamic node number determination formula in step S3. In a preferred embodiment, step S1 involves constructing and updating a machine learning model for quantifying the reputation of verification nodes and predicting their verification behavior based on the historical transaction verification records of the verification nodes. This includes constructing time-series feature data as input to the machine learning model, specifically: From the distributed ledger of the blockchain, retrieve the historical transaction verification records of all verification nodes within a preset time window; for each verification node, construct a time-series feature vector sequence from its historical transaction verification records; for each historical time slice, statistically analyze the feature data of the verification node within that time slice to form a time-series feature vector; arrange these vectors in chronological order to form the time-series feature vector sequence of the verification node; each time-series feature vector in the time-series feature vector sequence corresponds to a historical time slice, and this time-series feature vector contains the following multiple dimensions of feature data statistically obtained by the verification node within that historical time slice; the multiple dimensions of feature data specifically include: A1. Historical verification success rate, which represents the proportion of transactions that were successfully verified by the verification node within the historical time slice to the total number of verified transactions. A2. Historical average response latency represents the average time taken for the verification node to respond to and complete verification for all verification requests within the historical time slice. A3, Participation Frequency Distribution Vector, represents the participation frequency distribution of the verification node in verifying different subtypes of bone tumor data sharing transactions within this historical time slice. The subtypes include image data transactions, genomic data transactions, and pathology report data transactions.

[0006] In a preferred embodiment, the machine learning model is a recurrent neural network model based on an encoder-decoder architecture, and the recurrent neural network model contains multiple trainable weight parameters; the machine learning model analyzes historical transaction verification records and outputs a dynamic reputation score for each verification node and an evaluation result of the expected behavior of the current transaction, specifically including: The temporal feature vector sequence is input into the encoder part of the recurrent neural network model. The encoder encodes the sequence and outputs a hidden state vector that represents the long-term behavior pattern of the verification node. The decoder part of the recurrent neural network model performs the following operations based on the hidden state vector and the type characteristics of the current bone tumor data sharing transaction to be verified: B1. Predict the behavior tendency of the verification node for the current transaction and output a behavior prediction label. This behavior prediction label constitutes the expected behavior evaluation result. The value of the behavior prediction label includes three categories: reliable, risky, and unstable. B2. Calculate the dynamic reputation score of the verification node through a reputation fusion calculation process. The specific process of this reputation fusion calculation is as follows: B2.1. Perform a nonlinear transformation on the hidden state vector to obtain the basic reputation vector that represents the credibility of the historical behavior of the verification node itself. B2.2 Obtain the historical reputation scores of all neighboring verification nodes directly connected to the target verification node in the current reputation calculation in a preset verification node relationship network in the previous evaluation period, and determine the corresponding connection weights based on the historical interaction success rate between the target verification node and each neighboring verification node. Calculate a weighted average value based on the historical reputation scores and corresponding connection weights of all neighboring verification nodes as the network consensus reputation value. B2.3. Perform a first linear transformation on the basic reputation vector to obtain the self-behavioral reputation component; multiply the network consensus reputation value by a preset network influence adjustment parameter to obtain the network influence reputation component; add the self-behavioral reputation component and the network influence reputation component to obtain the original reputation value. B2.4 Input the original reputation value into an S-shaped growth function for nonlinear mapping, restrict the mapping output to between zero and one, and use the result as the dynamic reputation score of the target verification node at the current time.

[0007] In a preferred embodiment, in step S2, a dynamic two-layer trust network comprising a physical node layer and a logical organization layer is constructed. Specifically, the process is as follows: A dynamic two-layer trust network is constructed based on the affiliation between the verification node and its affiliated organization. The dynamic two-layer trust network consists of a physical node layer and a logical organization layer. Each physical node in the physical node layer corresponds to a verification node, and each physical node is assigned a node attribute. The value of this node attribute is the dynamic reputation score of the verification node corresponding to the physical node, obtained from step S1. Each logical organization node in the logical organization layer corresponds to an organization to which the verification node belongs. A connection edge is constructed for each pair of logical organization nodes in the logical organization layer, and this connection edge is assigned an edge weight attribute. The value of this edge weight attribute is calculated based on the historical cooperation data between the organizations corresponding to the pair of logical organization nodes in the organization attribute information. Then, graph embedding learning is performed on both the physical node layer and the logical organization layer. Through graph embedding learning, each physical node in the physical node layer is mapped to a first embedding vector of fixed length, and each logical organization node in the logical organization layer is mapped to a second embedding vector of fixed length.

[0008] In a preferred embodiment, the process of utilizing graph embedding technology and attention mechanisms, combined with the specific contextual features of the current transaction, to calculate the contextual trust level of each institution in this transaction involves: extracting the specific contextual features of the bone tumor data sharing transaction to be verified, including the transaction amount and data sensitivity level, and constructing a transaction context feature vector based on these features; then, using the second embedding vector of each logical institution node in the logical institution layer as the basic representation, introducing the transaction context feature vector as the global context, and dynamically calculating the association strength between any central logical institution node as the target and each of its neighboring logical institution nodes in the logical institution layer network under the context of this transaction through a multi-head attention mechanism. The process of calculating the association strength is as follows: The second embedding vector of the central logical institution node is multiplied by a first learnable parameter matrix to obtain the query representation; the second embedding vector of the neighboring logical institution nodes is multiplied by a second learnable parameter matrix to obtain the key representation; the dimensions of the first and second learnable parameter matrices are determined by the length of the second embedding vector and a preset attention space dimension; then, the scaled dot product of the query representation and the key representation is calculated, and the calculation results for all neighboring logical institution nodes are normalized exponentially to obtain the dynamic influence weight of the neighboring logical institution node on the central logical institution node; based on this, the contextual trust degree of the institution corresponding to the central logical institution node for the current transaction is calculated; the calculation process for this contextual trust degree is as follows: C1. For each neighboring logical institution node of the central logical institution node, fuse and encode the second embedding vector of the neighboring logical institution node with the historical cooperation data represented by the edge weight attribute of the edge connecting the two logical institution nodes to generate a neighbor contribution feature. Then multiply this neighbor contribution feature by the dynamic influence weight corresponding to the neighboring logical institution node to obtain its weighted contribution. Finally, sum the weighted contributions of all neighboring logical institution nodes to obtain the institution network trust component. C2. From the output of step S1, obtain the dynamic reputation scores of all verification nodes belonging to the central logical mechanism node, calculate the arithmetic mean of these dynamic reputation scores, and obtain the reputation components of the subordinate nodes. C3. Multiply the subordinate node reputation component by a preset institutional node trust transfer coefficient with a value range between zero and one, and then add it to the institutional network trust component to obtain an original trust value. C4. Input the original trust value into an S-shaped growth function for non-linear mapping, and restrict the mapping output value to between zero and one. The final output value is the contextual trust value of the institution corresponding to the central logical institution node in this transaction.

[0009] In a preferred embodiment, in step S3, based on the network average trust confidence level calculated from the contextual trust level and the transaction risk score extracted from the current transaction information and the data sensitivity level, the size of the dynamic verification node set is calculated using a dynamic node number determination formula. Specifically, this includes: First, calculating the network average trust confidence level. This calculation process involves: obtaining the contextual trust level of all institutions calculated in step S2 under this transaction, denoted as the first total number of institutions; calculating the arithmetic mean of these contextual trust levels, denoted as the first mean; and calculating the standard deviation of these contextual trust levels relative to the first mean, denoted as the first standard deviation; then, for each institution, calculating the fourth power of the difference between its contextual trust level and the first mean, obtaining the fourth power difference of that institution; calculating the product of the first total number and the fourth power of the first standard deviation, denoted as the first denominator; and so on. First, sum the fourth-order deviations of all institutions, divide the sum by the first total number to obtain the mean of the fourth-order deviations; then subtract the ratio of the mean of the fourth-order deviations to the fourth power of the first standard deviation from the sum of the first standard deviations. The result is the network average confidence level. Next, calculate the transaction risk score. The calculation process is as follows: extract the transaction amount and data sensitivity level from the current transaction information; add one to the transaction amount, take the natural logarithm, and multiply by a first weighting coefficient to obtain the amount risk component; square the data sensitivity level and multiply by a second weighting coefficient to obtain the sensitivity risk component; add the amount risk component and the sensitivity risk component, input the sum into a hyperbolic tangent function for mapping, and restrict the mapping result to between zero and one. This result is the transaction risk score. Finally, substitute the network average confidence level and the transaction risk score into the dynamic node number determination formula. The calculation process of the dynamic node number determination formula is as follows: E1. Calculate the difference between the number one and the network average confidence level. E2. Calculate the product of the transaction risk score and a preset risk sensitivity amplification coefficient, and then use the natural constant e as the base and this product as the exponent to calculate the exponential function value. E3. Multiply a preset scale adjustment coefficient, the difference calculated by E1, and the exponential function value calculated by E2 to obtain a scale adjustment amount. E4. Add a preset minimum number of safe nodes to the size adjustment amount to obtain the size calculation value; E5. Round the calculated size up to the nearest integer, which is the size of the dynamic verification node set.

[0010] In a preferred embodiment, the step of selecting verification nodes from the verification nodes to form a dynamic verification node set based on the contextualized trust level of the institution and the expected behavior assessment results of the verification nodes, and assigning verification tasks to the verification nodes in the dynamic verification node set, specifically includes: performing node selection and task assignment according to the size of the dynamic verification node set; firstly, allocating institutional quotas, specifically: for each institution, multiplying its contextualized trust level by the size of the dynamic verification node set, then dividing by the sum of the contextualized trust levels of all institutions, and rounding the result up, with the rounded value serving as the basis for the calculation. The organization obtains an initial number of quota nodes; then, within each organization, a two-step screening method is used to select the best nodes; finally, the verification nodes selected from all organizations using the two-step screening method constitute a dynamic verification node set; when assigning verification tasks to the verification nodes in the dynamic verification node set, a differentiated assignment strategy is adopted: for all verification nodes from the same organization that are selected into the dynamic verification node set, the verification node with the highest dynamic reputation score is selected and assigned a complete verification task; sampling verification tasks are assigned to the other verification nodes within the organization that are selected into the dynamic verification node set.

[0011] In a preferred embodiment, the specific process of collecting actual verification performance data in real time and comparing the actual verification performance data with the expected behavior evaluation result output by the machine learning model in step S1 in step S1 is as follows: During the verification of the current bone tumor data sharing transaction by the dynamic verification node set, the actual performance data of this verification task is collected in real time through deployed monitoring probes to form an actual performance data triplet; the actual performance data triplet consists of the following three indicators: U1, Actual consensus time, is the length of time from the start of verification at the first verification node to the completion of consensus. U2. Actual node response consistency, which is the proportion of the number of nodes that output the same correct verification result in the set of dynamic verification nodes to the total number of nodes in the set. U3, Average Response Time of Actual Nodes, is the average time taken by all verification nodes in the dynamic verification node set from receiving the verification task to submitting the verification result. Simultaneously, data relevant to this verification task is extracted from the expected behavior evaluation results and dynamic reputation scores output by the machine learning model in step S1, serving as expected performance data to form expected performance data triplets; the expected performance data triplets consist of the following three estimation metrics: P1, Expected consensus time, is estimated based on the predicted behavior labels of each verification node selected into the dynamic verification node set and its historical average response delay. P2. Expected node response consistency is estimated based on the proportion of the number of verification nodes whose predicted behavior label is marked as "reliable" among the verification nodes selected into the dynamic verification node set to the total number of nodes in the set. P3, Expected average node response latency, is estimated based on the historical average response latency of each verification node selected in the dynamic verification node set; Next, each indicator in the actual performance data triplet is compared one by one with the corresponding estimated indicator in the expected performance data triplet to calculate the relative deviation between each pair of indicators. Then, a dynamic weight is assigned to the relative deviation of each type of performance indicator. Finally, the relative deviation of each type of performance indicator is multiplied by a preset deviation sensitivity coefficient and then input into an S-shaped growth function for nonlinear compression mapping. The mapping result is multiplied by the corresponding dynamic weight, and the results of all three types of performance indicators after weighted processing are summed to obtain a comprehensive difference degree.

[0012] In a preferred embodiment, the step of performing a feedback adjustment process based on the differences generated by the comparison specifically includes: model parameter adjustment for the machine learning model in step S1, specifically as follows: First, multiply the comprehensive difference degree by a preset model adjustment sensitivity coefficient, and then input the multiplication result into a hyperbolic tangent function for mapping to obtain a value between negative one and positive one, which serves as a scaling factor for model parameter updates; Second, using a recent batch of data containing actual validation performance labels, calculate the gradient of the current trainable weight parameters of the machine learning model relative to an adaptive loss function; Then, multiply a preset meta-learning rate used to control the overall update step size of the model parameters by the scaling factor and the gradient of the loss function to obtain the update amount of the trainable weight parameters of the machine learning model; Finally, adjust the trainable weight parameters of the machine learning model in step S1 based on this update amount; The decision parameter adjustment is performed on the scale adjustment coefficient included in the dynamic node number determination formula in step S3, specifically as follows: R1. Based on the consistency between the actual consensus time and the actual node response, estimate an ideal number of verification nodes under the current balance between efficiency and security. R2. Calculate a direction discrimination value, which is equal to the actual size of the dynamic verification node set determined in step S3 minus the ideal number of verification nodes, and then divided by the ideal number of verification nodes; take the sign function of the direction discrimination value to obtain a direction signal; R3. Calculate an adjustment basis consisting of immediate difference and historical difference; the immediate difference component is the product of the comprehensive difference degree of the current round and a preset immediate difference weighting factor; the historical difference component is the comprehensive difference degree of the past several rounds, which is multiplied by a decay weight with a preset historical error decay factor as the base and the time difference from the current round as the exponent, then summed, and finally multiplied by the difference between the number one and the immediate difference weighting factor; add the immediate difference component and the historical difference component to obtain the adjustment basis; R4. Multiply the direction signal, a preset adjustment intensity coefficient for controlling the overall amplitude, and the adjustment basis to obtain the adjustment amount of the scale adjustment coefficient. R5. Based on this adjustment amount, within the preset lower and upper limits of the scale adjustment coefficient, update the scale adjustment coefficient in the dynamic node number determination formula described in step S3.

[0013] This application also provides a blockchain communication system for secure sharing of bone tumor data, specifically including: a node reputation and behavior prediction module, an institutional trust dynamic assessment module, a dynamic verification networking and decision-making module, and a real-time monitoring and feedback adjustment module, wherein; Node Reputation and Behavior Prediction Module: When a bone tumor data sharing transaction verification request is received, the module builds and updates a machine learning model based on the historical transaction verification records of the verification node to output the dynamic reputation score of each verification node and the expected behavior assessment result of the current transaction. Institutional Trust Dynamic Assessment Module: This module is used to construct a two-layer trust network based on the dynamic reputation score, the attribute information of the institution to which the verification node belongs, and the current transaction context, and to calculate the contextual trust level of each institution in this transaction by combining an attention mechanism. Dynamic verification network and decision-making module: It is used to integrate the expected behavior evaluation results and contextual trust level, select verification nodes from the verification nodes to form a dynamic verification node set and assign differentiated verification tasks. The size of the set is calculated by a dynamic node number determination formula containing an adjustable size adjustment coefficient. The input of the formula includes the network average trust level confidence obtained based on the contextual trust level and the transaction risk score calculated based on the current transaction information. Real-time monitoring and feedback adjustment module: When the dynamic verification node set performs verification, it collects the actual verification performance data and compares it with the expected behavior evaluation results, and adjusts the machine learning model and the scale adjustment coefficient according to the difference feedback.

[0014] The beneficial effects of this invention are as follows: By constructing a dynamic evaluation system that integrates node behavior prediction and institutional contextual trust, the scheme achieves accuracy and foresight in the selection of verification nodes, significantly reduces verification latency, and its elastic verification scale mechanism based on trust status and transaction risk effectively avoids resource redundancy and waste, significantly improves consensus efficiency, and introduces real-time performance monitoring and dual-loop feedback adjustment, enabling the system to have continuous self-optimization capabilities and dynamically adapt to changes in node behavior and network fluctuations. Thus, while ensuring high security of bone tumor data sharing transactions, it achieves efficient and reliable intelligent collaborative verification. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a block diagram of the system structure of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0018] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0019] Example 1 This embodiment provides, for example Figure 1The illustrated blockchain communication method for secure sharing of bone tumor data includes the following steps: Step S1: In response to receiving a bone tumor data sharing transaction verification request in the blockchain network, based on the historical transaction verification records of the verification nodes, construct and update a machine learning model for quantifying the reputation of the verification nodes and predicting their verification behavior; the historical transaction verification records include the historical verification success rate, historical average response latency, and participation frequency distribution of each verification node in the verification of different subtypes of bone tumor data transactions; the machine learning model analyzes the historical transaction verification records and outputs the dynamic reputation score of each verification node and the expected behavior evaluation result for the current transaction; Step S2: Based on the dynamic reputation score of the verification node, the attribute information of the institution to which the verification node belongs, and the specific contextual features of the current transaction, a dynamic two-layer trust network is generated; the institution attribute information includes historical cooperation data between institutions; the physical node layer of the two-layer trust network corresponds to each verification node, and the logical institution layer corresponds to the institution to which each verification node belongs; using graph embedding technology and attention mechanism, combined with the specific contextual features of the current transaction, the contextual trust degree of each institution in this transaction is calculated. Step S3: Based on the expected behavior assessment results of the verification nodes and the contextual trust level of the institution, execute optimization decisions. Optimization decisions include: selecting verification nodes from the verification nodes to form a dynamic verification node set based on the contextual trust level of the institution and the expected behavior assessment results of the verification nodes, and assigning verification tasks to the verification nodes in the dynamic verification node set. The size of the dynamic verification node set is calculated using a dynamic node number determination formula based on the network average trust confidence level calculated from the contextual trust level and the transaction risk score calculated by extracting the transaction amount and data sensitivity level from the current transaction information. The dynamic node number determination formula includes an adjustable size adjustment coefficient. Step S4: During the verification process of the dynamic verification node set, collect actual verification performance data in real time. The actual verification performance data includes consensus achievement time, node response consistency, and node response latency. Compare the actual verification performance data with the expected behavior evaluation results output by the machine learning model in step S1. Based on the differences generated by the comparison, execute a feedback adjustment process. The feedback adjustment process is used to adjust the machine learning model in step S1 and to adjust the scale adjustment coefficient included in the dynamic node number determination formula in step S3.

[0020] In this embodiment, it is specifically necessary to explain step S1, which involves constructing and updating a machine learning model for quantifying the reputation of verification nodes and predicting their verification behavior based on the historical transaction verification records of the verification nodes. This includes constructing time-series feature data as input to the machine learning model. Specifically, this involves retrieving the historical transaction verification records of all verification nodes within a preset time window from the distributed ledger of the blockchain. The preset time window can be configured according to the actual scenario, such as the last 30 days or the last 1000 block heights, to ensure the timeliness of the data. The historical transaction verification records are extracted from the verification event logs recorded in each block and include at least the unique transaction identifier, the identifier of the node participating in the verification, the verification result (success / failure), and the verification time. For each verification node, its historical transaction verification records are constructed into a time-series feature vector sequence. The construction method of the time-series feature vector sequence is as follows: the preset time window is divided into several consecutive historical time slices (for example, each time slice is 1 day long). For each historical time slice, the feature data of the verification node within that time slice is statistically analyzed to form a time-series feature vector. These vectors are arranged in chronological order to form the time-series feature vector sequence of the verification node. Each time-series feature vector in the time-series feature vector sequence corresponds to a historical time slice. The time-series feature vector contains the following multiple dimensions of feature data statistically obtained by the verification node within that historical time slice. The multiple dimensions of feature data specifically include: A1. Historical Verification Success Rate: This represents the proportion of successfully verified transactions completed by the verification node within a given historical time slice, out of the total number of verified transactions. The formula is: the number of successfully verified transactions within the time slice divided by the total number of transactions participating in verification within the time slice. This characteristic directly reflects the reliability of the verification node. A2. Historical Average Response Latency: This represents the average time taken by the verification node to respond to and complete verification for all verification requests within a given historical time slice. The calculation method is: for all verification transactions in which the verification node participated within the time slice, calculate the time difference between receiving the verification task and submitting the verification result, then calculate the arithmetic mean. This characteristic reflects the reliability of the verification node. Processing efficiency; A3, Participation frequency distribution vector, represents the participation frequency distribution of the verification node in verifying different subtypes of bone tumor data sharing transactions within the historical time slice. The subtypes include at least image data transactions, genomic data transactions, and pathology report data transactions. The calculation process of the participation frequency distribution vector is as follows: First, count the number of times the verification node participates in the verification of each subtype of transaction within the time slice. Then, divide the number of participations in each subtype by the total number of verification participations of the verification node within the time slice to obtain a normalized frequency distribution vector. This feature is used to characterize the verification behavior preferences of the verification node and helps the model identify the node's expertise or behavioral pattern differences in different data types. The machine learning model is a recurrent neural network model based on an encoder-decoder architecture. This recurrent neural network model contains multiple trainable weight parameters, which are determined through an optimization algorithm during model training and dynamically adjusted based on the comprehensive difference during the feedback adjustment process in step S4. Specifically, a long short-term memory network or a gated recurrent unit can be used. The encoder encodes the input temporal feature vector sequence, capturing the patterns of node behavior changes over time. The decoder uses the contextual information output by the encoder (i.e., the hidden state vector) and the contextual features of the current transaction (e.g., one-hot encoding of the transaction type) to perform behavior classification and reputation score calculation. The machine learning model analyzes historical transaction verification records and outputs a dynamic reputation score for each verification node and an evaluation result of the expected behavior of the current transaction, specifically including: The temporal feature vector sequence is input into the encoder part of the recurrent neural network model. The encoder encodes the sequence and outputs a hidden state vector representing the long-term behavior pattern of the verification node. The decoder part of the recurrent neural network model performs the following operations based on the hidden state vector and the type features of the current bone tumor data sharing transaction to be verified: B1. Predict the behavior tendency of the verification node towards the current transaction and output a behavior prediction label. This behavior prediction label constitutes the expected behavior evaluation result. The value of the behavior prediction label includes three categories: reliable, risky, and unstable. Specifically, the decoder may contain a fully connected layer connected to a Softmax classifier, which fuses the hidden state vector and the current transaction type features into a vector. The mapping is represented by probability distributions for three categories. The category with the highest probability is taken as the output behavior prediction label. The "Reliable" label indicates that the node is predicted to complete the verification correctly and in a timely manner; the "Risk" label indicates that the node may delay or give an incorrect verification result; and the "Unstable" label indicates that the node's behavior is difficult to predict. B2. The dynamic reputation score of the verification node is calculated through a reputation fusion calculation process. The specific process of this reputation fusion calculation process is as follows: B2.1. A nonlinear transformation is performed on the hidden state vector to obtain the basic reputation vector representing the credibility of the verification node's own historical behavior. The nonlinear transformation can be achieved by using one or more fully connected layers in conjunction with an activation function (such as ReLU). The purpose is to integrate the comprehensive data extracted by the encoder. The temporal feature is mapped to a low-dimensional vector that can be directly used for reputation calculation; B2.2, obtain the historical reputation scores of all neighboring verification nodes directly connected to the target verification node for the current reputation calculation in a pre-defined verification node relationship network in the previous evaluation period, and determine the corresponding connection weights based on the historical interaction success rate between the target verification node and each neighboring verification node. A weighted average is calculated based on the historical reputation scores of all neighboring verification nodes and their corresponding connection weights, serving as the network consensus reputation value; the verification node relationship network is a pre-constructed graph structure, where nodes represent verification nodes and edges represent historical joint verification or communication relationships between nodes; connection weights can be determined based on the historical joint verification relationships between two nodes. The proportion of consistent results during transactions is used for calculation; historical reputation scores are derived from the results calculated and stored in the previous round of step S1, which reflects the propagation and aggregation of reputation in the network; B2.3, perform a first linear transformation on the basic reputation vector to obtain the self-behavioral reputation component; multiply the network consensus reputation value by a preset network influence adjustment parameter to obtain the network influence reputation component; where the network influence adjustment parameter is an adjustable factor between zero and one, used to control the influence strength of the network consensus reputation value on the final reputation score; the network influence adjustment parameter is learned together with other parameters during the model training phase, or preset by the system administrator according to the needs of network consensus strength, for example, it can be initially set to 0.5. Add the self-behavioral reputation component to the network influence reputation component to obtain the original reputation value; this addition operation realizes the fusion of the node's own behavioral performance and the local network community reputation evaluation; B2.4. Input the original reputation value into a sigmoid function for nonlinear mapping, restrict the mapping output to between zero and one, and use this result as the dynamic reputation score of the target verification node at the current moment. The sigmoid function maps the original reputation value (an arbitrary real number) to the (0,1) interval as a standardized reputation score. The closer the score is to 1, the higher the reputation; the closer it is to 0, the lower the reputation. The core innovation of the entire reputation fusion calculation process lies in its innovative approach. It not only relies on the historical behavioral characteristics of the verification node itself (reflected in the basic reputation vector), but also creatively introduces a weighted feedback of the historical reputation of its local network neighbors (reflected in the network consensus reputation value). Furthermore, it balances the contributions of both through network influence adjustment parameters. This makes the reputation score not only a reflection of individual behavior but also incorporates network topology and community consensus information. This allows for more robust and earlier identification of unstable nodes exhibiting abnormal behavior or attempting to integrate into reliable communities. Compared to methods that only use their own historical statistics (such as mean and variance), it possesses stronger anomaly detection capabilities and robustness.

[0021] In this embodiment, it is particularly important to explain step S2, which involves constructing a dynamic two-layer trust network comprising a physical node layer and a logical mechanism layer. The specific process is as follows: Based on the hierarchical relationship between verification nodes and their affiliated institutions, a dynamic two-layer trust network is constructed. This construction is achieved through a two-layer graph structure: the first layer is the physical node layer, whose node set consists of all physical server or client processes participating in verification (i.e., verification nodes); the second layer is the logical institution layer, whose node set consists of the entities such as hospitals, research laboratories, or companies to which these verification nodes belong. The connection between the two layers is uniquely determined by the hierarchical relationship between the verification node and the institution; that is, a logical institution node is connected to all its subordinate physical verification nodes. The dynamic two-layer trust network consists of the physical node layer and the logical institution layer. Each physical node in the management node layer corresponds to a verification node, and each physical node is assigned a node attribute. The value of this node attribute is the dynamic reputation score of the verification node corresponding to the physical node, obtained from step S1. Each logical organization node in the logical organization layer corresponds to an organization to which the verification node belongs. A connection edge is constructed for each pair of logical organization nodes in the logical organization layer, and an edge weight attribute is assigned to this connection edge. The value of this edge weight attribute is calculated based on the historical cooperation data between the organizations corresponding to the pair of logical organization nodes in the organization attribute information. The historical cooperation data between organizations is quantified as the cooperation between the two corresponding organizations within a set period in the past. The number of successful data sharing transactions or the total amount of transactions completed can be used as edge weights. For example, the number of transactions can be normalized to the [0,1] interval, or the logarithm of the transaction amount can be normalized to a minimum value. This edge weight attribute is used to characterize the historical cooperation closeness between institutions. Then, graph embedding learning is performed on the physical node layer and the logical institution layer respectively. Graph embedding learning can use algorithms such as Node2Vec and GraphSAGE, which aim to map the nodes in the graph structure (including physical nodes and logical institution nodes) to a low-dimensional continuous vector space, while preserving the structural similarity between nodes (such as the node vectors of the same institution should be close) and attribute information (such as physical...). (Node reputation score); Through graph embedding learning, each physical node in the physical node layer is mapped to a fixed-length first embedding vector, and each logical mechanism node in the logical mechanism layer is mapped to a fixed-length second embedding vector. The first embedding vector is used to represent the relationship and features of the corresponding verification node in the physical node layer network structure in a low-dimensional space, and the second embedding vector is used to represent the relationship and features of the corresponding mechanism in the logical mechanism layer network structure in a low-dimensional space. This step transforms discrete, high-dimensional graph relationship data into continuous, low-dimensional numerical vectors, providing a unified and processable data input for subsequent complex computations based on neural networks. By utilizing graph embedding technology and attention mechanisms, and combining the specific contextual features of the current transaction, the contextual trust level of each institution in this transaction is calculated. The specific process is as follows: First, the specific contextual features of the bone tumor data sharing transaction to be verified are extracted. These features include at least the transaction amount and the data sensitivity level. The data sensitivity level can be predefined based on factors such as whether the data contains personally identifiable information or whether it represents key research findings. For example, it can be divided into levels such as public, internal, secret, and top secret, and encoded numerically. Based on these features, a transaction context feature vector is constructed. Next, using the second embedding vector of each logical institution node in the logical institution layer as the basic representation, the transaction context is introduced. The feature vector serves as the global context. Through a multi-head attention mechanism, the association strength between any central logical institution node (the target node) and each of its neighboring logical institution nodes in the logical institution layer network under the current transaction context is dynamically calculated. The association strength calculation process is as follows: multiplying the second embedding vector of the central logical institution node by a first learnable parameter matrix yields the query representation; multiplying the second embedding vectors of the neighboring logical institution nodes by a second learnable parameter matrix yields the key representation. The dimensions of the first and second learnable parameter matrices are jointly determined by the length of the second embedding vector and a preset attention space dimension. For example, if the length of the second embedding vector is... The preset attention space dimension is Then the dimensions of the first learnable parameter matrix and the second learnable parameter matrix are both . Preset attention space dimensions It is usually set to the second embedding vector length. The values ​​are 1 / 4, 1 / 2, etc., to reduce computation and prevent overfitting; then the scaled dot product of the query representation and the key representation is calculated, which is to multiply the query representation vector by the transpose of the key representation vector and then divide by the attention space dimension. The gradient is scaled using the square root to stabilize it. Then, the calculation results of all neighboring logical institution nodes are normalized exponentially to obtain the dynamic influence weight of the neighboring logical institution node on the central logical institution node. The normalization exponential processing is the Softmax function, which ensures that the sum of the dynamic influence weights of all neighboring logical institution nodes on the same central logical institution node is 1. This dynamic influence weight reflects the relative importance of different neighboring institutions' trust in the target institution under the current specific transaction characteristics, realizing the "contextual awareness" of trust assessment, rather than static global weights. Based on this, the contextual trust degree of the institution corresponding to the central logical institution node for the current transaction is calculated. The calculation process of this contextual trust degree is as follows: C1. For each neighboring logical institution node of the central logical institution node, the second embedding vector of the neighboring logical institution node is fused and encoded with the historical cooperation data represented by the edge weight attribute of the edge connecting the two logical institution nodes. The fusion encoding is implemented through a multilayer perceptron to generate a neighbor contribution feature. Then, this neighbor contribution feature is multiplied by the dynamic influence weight corresponding to the neighboring logical institution node to obtain its weighted contribution. Finally, the weighted contributions of all neighboring logical institution nodes are summed to obtain the institutional network trust component. The essence of this step is: from all neighboring machines of the target institution Starting from the structure, combining the historical cooperation closeness with them (edge ​​weights) and their feature representation in the network (second embedding vector), and assigning different weights according to the current transaction context, a comprehensive metric representing "trust endorsement from the partner community" is aggregated; C2, from the output of step S1, obtain the dynamic reputation scores of all verification nodes belonging to the central logical institution node, calculate the arithmetic mean of these dynamic reputation scores, and obtain the subordinate node reputation component; this component directly reflects the average reliability and efficiency of the institution's currently schedulable verification resources, and is a manifestation of the institution's own real-time capabilities; C3, multiply the subordinate node reputation component by a preset institution node trust transfer coefficient with a value range between zero and one, and then combine it with the machine The network trust components are summed to obtain an initial trust value. The institutional node trust transfer coefficient is used to adjust the contribution ratio of subordinate node reputation components in the initial trust value calculation. This coefficient can be preset based on the analysis of the correlation between institutional reputation and node reputation in historical data; for example, it can be set to 0.6, indicating that in the initial trust value, the institution's own node reputation contributes 60%, and the trust contribution from the partner network accounts for 40%. This institutional node trust transfer coefficient provides flexibility for strategy adjustment. C4. The initial trust value is input into an S-shaped growth function for non-linear mapping, and the mapped output value is restricted to between zero and one. This final output value is the value of the institution corresponding to the central logical institutional node in this transaction. Contextualized trust score, with its S-shaped growth function typically referring to the Sigmoid function, maps any real number to the (0,1) interval as a standardized probability or confidence output. The effect of the entire contextualized trust score calculation process is that it does not simply perform a static scoring of the institution, but rather places the institution in a dynamic system composed of historical cooperation networks, current transaction context, and real-time subordinate node states for evaluation. It achieves dynamic allocation of evaluation weights through an attention mechanism and aggregates multi-dimensional information through fusion encoding. The final output contextualized trust score is a highly specific indicator that accurately reflects the institution's trust status "in this transaction, at this moment," providing a crucial basis for refined decision-making in subsequent steps.

[0022] In this embodiment, it is specifically necessary to explain step S3, which involves calculating the size of the dynamic verification node set based on the network average trust confidence level calculated from the contextualized trust level and the transaction risk score extracted from the current transaction information and the data sensitivity level. This is achieved through a dynamic node number determination formula. Specifically, this includes: First, the network average trust confidence score is calculated. This calculation process involves: obtaining the contextualized trust scores of all institutions in this transaction, calculated in step S2, and denoteing the total number of institutions as the first total. The arithmetic mean of these contextualized trust scores is then calculated and denoted as the first mean. The standard deviation of these contextualized trust scores relative to the first mean is also calculated and denoted as the first standard deviation. Then, for each institution, the fourth power of the difference between its contextualized trust score and the first mean is calculated to obtain the fourth power deviation value for that institution. The product of the first total and the fourth power of the first standard deviation is calculated and denoted as the first denominator. The sum of the fourth power deviation values ​​for all institutions is calculated, and the sum is divided by the first total to obtain the mean of the fourth power deviation values. Finally, the ratio of the mean of the fourth power deviation values ​​to the fourth power of the first standard deviation is subtracted from the first value. The result is the network average trust confidence score. Essentially, this calculation method assesses the concentration of trust by measuring the kurtosis (the ratio of the fourth central moment to the fourth power of the standard deviation) of the contextualized trust score distribution. The more concentrated the distribution (i.e., the trust levels of each institution are close to each other), the higher the calculated average trust level of the network, indicating that the network consensus foundation is more solid.For example, if the contextualized trust levels of the three institutions are 0.8, 0.85, and 0.9 respectively, their distribution is concentrated, and the calculated confidence level may be as high as 0.95; if they are 0.5, 0.8, and 0.95, their distribution is dispersed, and the confidence level may be as low as 0.7. The closer the network's average trust level is to 1, the closer the trust levels of the institutions are, the more stable the network's trust state, the stronger the consensus foundation, and the fewer verification nodes are required. Secondly, a transaction risk score is calculated. This calculation process involves extracting the transaction amount and data sensitivity level from the current transaction information. For example, the data sensitivity level can be predefined as levels 1 to 5, where level 1 is public data and level 5 is highly sensitive data containing personally identifiable information and key research results. The data sensitivity level is a preset numerical level. The transaction amount is added by one and the natural logarithm is taken, then multiplied by a first weighting coefficient to obtain the amount risk component. The "adding by one and taking the natural logarithm" process aims to reduce the linear risk growth brought about by huge transactions, making it present as a logarithm. The increase is more in line with the actual risk assessment logic. The first weighting coefficient can be used to adjust the contribution of amount to risk, for example, it can be preset to 0.3; the data sensitivity level is squared and then multiplied by a second weighting coefficient to obtain the sensitivity risk component; "squaring" is to amplify the risk brought by the high sensitivity level, because the risk increment brought by the data sensitivity from level 4 to level 5 is much greater than that from level 1 to level 2; the second weighting coefficient can be used to adjust the contribution of sensitivity to risk, for example, it can be preset to 0.7 to emphasize the core position of data sensitivity in medical data sharing; the amount risk component and the sensitivity risk component are added together, and the sum is input into a hyperbolic tangent function for mapping, and the mapping result is restricted to the range between zero and one. This result is the transaction risk score; the hyperbolic tangent function (tanh) maps any real number to (-1,1). Since the amount risk component and the sensitivity risk component are both non-negative, their sum after tanh mapping falls in the range of (0,1), which meets the score definition. Finally, the network average trust confidence score and transaction risk score are substituted into the dynamic node number determination formula. The calculation process of this dynamic node number determination formula is as follows: E1, calculate the difference between the number 1 and the network average trust confidence score; this difference (1 - network average trust confidence score) directly reflects the dispersion or instability of network trust. The larger the difference, the more nodes are needed to reach a consensus; E2, calculate the product of the transaction risk score and a preset risk sensitivity amplification coefficient, and then use the natural constant e as the base and this product as the exponent to calculate the exponential function value; the risk sensitivity amplification coefficient is used to control the risk impact on nodes. The amplification effect of point growth, for example, can be preset to 2.0. The exponential function exp(risk sensitivity amplification coefficient * transaction risk score) makes the node size respond exponentially to high-risk transactions, greatly enhancing the security of high-value or high-sensitivity transactions; E3, multiply a preset size adjustment coefficient, the difference calculated by E1, and the exponential function value calculated by E2 to obtain a size adjustment amount; the size adjustment coefficient is the benchmark parameter controlling the overall adjustment range, for example, it can be preset to 5.0. This step multiplies the trust dispersion with the risk amplification effect, reflecting that "risk increases with low trust..." The collaborative decision-making logic of "greater harm in the environment"; E4, add a preset minimum number of safe nodes to the size adjustment amount to obtain the size calculation value; the minimum number of safe nodes is the basic security configuration of the system, used to prevent the minimum collusion attack, for example, it can be set to 4; E5, round up the size calculation value, the integer obtained is the size of the dynamic verification node set; rounding up ensures that the number of nodes is an integer; among them, the minimum number of safe nodes is the lower limit of the number of nodes used to prevent collusion attacks, the size adjustment coefficient is used to control the benchmark amplitude of the influence of trust and risk factors on the number of nodes, and the risk sensitivity amplification coefficient is used to... The sensitivity of the transaction risk score to the impact of the number of nodes is controlled. The effect of the dynamic node number determination formula is that it is not a fixed value, nor is it a simple linear addition. Instead, it creatively couples the network average trust confidence, which represents the global state of the network, with the transaction risk score, which represents the risk of the current individual transaction, through multiplication and an exponential function. This allows the verification scale to adapt to the overall trust health of the network (expanding when discrete) and to respond extremely sensitively to the risk of a single transaction (exponential expansion when high risk), thus achieving refined, intelligent, and dynamic allocation of security resources. Based on the contextual trust level of the organization and the expected behavior assessment results of the verification nodes, verification nodes are selected from the verification nodes to form a dynamic verification node set, and verification tasks are assigned to the verification nodes in the dynamic verification node set. Specifically, this includes: performing node selection and task assignment according to the size of the dynamic verification node set; first, allocating quotas to organizations, specifically: for each organization, multiply its contextual trust level by the size of the dynamic verification node set, then divide by the sum of the contextual trust levels of all organizations, and round up the result. This rounded value is the initial quota number of nodes obtained by the organization. This step ensures that high-trust organizations obtain more seats in the verification group, reflecting the principles of "equal rights and responsibilities" and "priority of superior resources". Due to the use of rounding up, the sum of the quotas of each organization may be slightly larger than the original size, but this is within a controllable range and tends to be safer; then, within each organization, a two-step selection method is used to select the best nodes within the organization, specifically: from the organization's subordinate... Of all the verification nodes, based on the expected behavior assessment results output in step S1, verification nodes with the behavior prediction label "reliable" are selected first. If the number of verified nodes with the behavior prediction label "reliable" is greater than or equal to the organization's initial quota of nodes, then the verification nodes with the highest dynamic reputation score and the number equal to the initial quota of nodes are selected. If the number of verified nodes with the behavior prediction label "reliable" is less than the organization's initial quota of nodes, then all of these verification nodes are selected first, and then verification nodes are selected from the other verification nodes under the organization that have not been selected, in descending order of their dynamic reputation scores, until the total number of selected verification nodes reaches the organization's initial quota of nodes. The effect of this "two-step screening method" is that the first priority is the immediate behavior reliability of the node (guaranteed by the "behavior prediction label" in step S1), and on the basis of meeting the reliability standard, the node with the best historical comprehensive performance (highest dynamic reputation score) is selected. This ensures the reliability of the core members of the verification team and makes full use of the reputation mechanism to select high-quality nodes; ultimately, the verification nodes selected by all institutions according to the two-step screening method together constitute a dynamic set of verification nodes. When assigning verification tasks to verification nodes in the dynamic verification node set, a differentiated assignment strategy is adopted: For all verification nodes from the same institution selected into the dynamic verification node set, the verification node with the highest dynamic reputation score is selected and assigned a full verification task, which requires verifying all transactions within the block; for the other verification nodes within the same institution selected into the dynamic verification node set, sampling verification tasks are assigned, which require verifying only a randomly selected proportion of transactions within the block; the proportion of transactions verified by the sampling verification task can be dynamically adjusted according to the network average trust confidence level; for example, when the network average trust confidence level is high (e.g., greater than 0.9), it indicates a good network trust status, and the sampling proportion can be set to a lower 2. To improve efficiency, the sampling rate is 0%. When the confidence level is low (e.g., less than 0.7), it indicates uncertainty in the network, and the sampling ratio can be increased to 40% to enhance the verification strength. This "master-slave verification + dynamic sampling" task assignment mode ensures that at least one high-reputation node performs global oversight (complete verification). By randomly sampling the remaining nodes, a cross-validation network is formed, which can effectively detect malicious behavior and significantly reduce the overall computation and communication overhead. It is especially suitable for scenarios with certain requirements for processing timeliness, such as bone tumor data sharing. The entire screening and assignment process deeply integrates information from three dimensions: institutional trust (macro), node behavior prediction (micro-level immediate), and node historical reputation (micro-level long-term). The constructed dynamic verification node set has both structural rationality and individual quality.

[0023] In this embodiment, the specific process of collecting actual verification performance data in real time and comparing the actual verification performance data with the expected behavior evaluation result output by the machine learning model in step S1 in step S4 is as follows: During the verification process of the current bone tumor data sharing transaction executed by the dynamic verification node set, the actual performance data of this verification task is collected in real time through the deployed monitoring probe, forming an actual performance data triplet; the actual performance data triplet consists of the following three indicators: U1, actual consensus time, which is the length of time from the start of verification by the first verification node to the completion of consensus; consensus completion can be defined as the moment when more than two-thirds of the verification nodes reach a consensus and complete the verification result on the chain; U 2. Actual node response consistency: This value represents the proportion of nodes in the dynamic verification node set that output the same correct verification result to the total number of nodes in the set. This proportion is calculated by dividing the number of nodes outputting the same correct verification result by the total number of nodes in the dynamic verification node set. The determination of the same correct verification result is made by the smart contract according to the preset business logic. 3. Actual node average response latency: This value represents the average time taken by all verification nodes in the dynamic verification node set from receiving the verification task to submitting the verification result. This average is calculated by summing the time taken by each verification node in the dynamic verification node set from receiving the task to submitting the result, and then dividing by the total number of nodes in the dynamic verification node set. Simultaneously, data relevant to this verification task is extracted from the expected behavior evaluation results and dynamic reputation scores output by the machine learning model in step S1, forming expected performance data triplets. These triplets consist of the following three estimation metrics: P1, Expected consensus time, estimated based on the predicted behavior labels and historical average response delays of each verification node in the selected dynamic verification node set; for example, the maximum historical average response delay of nodes with the predicted behavior label "reliable" is used as a conservative estimate of the expected consensus time; P2, Expected node response consistency, estimated based on the proportion of verification nodes with the predicted behavior label "reliable" to the total number of nodes in the selected dynamic verification node set; this estimation is based on the reasonable assumption that nodes predicted as "reliable" will output correct verification results with a very high probability; P3, Expected average node response delay, estimated based on the historical average response delay of each verification node in the selected dynamic verification node set; specifically, this is achieved by calculating the arithmetic mean of the historical average response delays of these verification nodes; this estimation reflects the average response speed of the verification node set under historical normal conditions. Next, each indicator in the actual performance data triplet is compared one by one with the corresponding estimated indicator in the expected performance data triplet, and the relative deviation between each pair of indicators is calculated. The calculation process for the relative deviation is as follows: the absolute value of the difference between the actual indicator value and the expected estimated indicator value is obtained, and then divided by the larger value between the expected estimated indicator value and a very small positive number. The very small positive number is used to prevent the denominator from being zero when calculating the relative deviation. The very small positive number is usually set to a very small value, such as 1e-5. Then, a dynamic weight is assigned to the relative deviation of each type of performance indicator. The calculation process for the dynamic weight is as follows: first, the average relative deviation of each type of performance indicator within a preset historical window period is calculated, and then a normalized exponential function is applied to these average relative deviations to make the sum of the dynamic weights of all types of performance indicators equal to one, and performance indicators with larger historical average relative deviations are assigned higher dynamic weights. The normalized exponential function is the Softmax function. For example, assuming the historical window represents the most recent 10 rounds of validation, the average relative deviation of each metric is calculated over these 10 rounds. Then, the weight is calculated using the Softmax function. If the historical deviation of "actual node response consistency" remains large, its dynamic weight in the current round will automatically increase, making the system more attentive to prediction errors in this metric. Finally, the relative deviation of each performance metric is multiplied by a preset deviation sensitivity coefficient and then input into an S-shaped growth function for nonlinear compression mapping. The S-shaped growth function is an exponential growth function used to map the input value to the interval between zero and one. The sensitivity coefficient is used to adjust the saturation rate of the indicator after the deviation enters the Sigmoid function, for example, it can be preset to 2.0; the mapping result is multiplied by the corresponding dynamic weight, and the results of all three types of performance indicators after weighting are summed to obtain a comprehensive difference degree; the comprehensive difference degree is a scalar value between 0 and 3 (because there are at most three types of indicators, and the maximum contribution of each type after weighting is 1). The closer its value is to 0, the higher the actual performance is to the prediction; the larger the value, the more serious the deviation. The comprehensive difference degree is used to quantify the overall degree of deviation between the actual validation performance and the model prediction in this round. Based on the differences generated by the comparison, a feedback adjustment process is performed, specifically including: model parameter adjustment for the machine learning model in step S1, specifically: First, the overall difference is multiplied by a preset model adjustment sensitivity coefficient, and then the result is input into a hyperbolic tangent function for mapping, obtaining a value between -1 and positive 1, which serves as the scaling factor for model parameter updates; the model adjustment sensitivity coefficient is used to control the slope of the influence of the overall difference on the scaling factor, for example, it can be set to 2.0. The hyperbolic tangent function (tanh) maps the input to (-1,1), making the scaling factor both positive and negative, thus theoretically supporting bidirectional adjustment of model parameters. When the overall difference is large, the absolute value of the scaling factor is close to 1, and the update step size is large; when the difference is small, the scaling factor is close to 0, and the update step size is small, achieving adaptive step size control; Second, using a recent batch of data containing actual validation performance labels, the gradient of the current trainable weight parameters of the machine learning model with respect to an adaptive loss function is calculated; the recent batch of data may include the data from the last 20 validation tasks. The actual behavior of a node (e.g., whether it is delayed, whether the verification result is correct) is used as the true label. The adaptive loss function can be the cross-entropy loss function, used to measure the difference between the model's predicted behavior label and the true label. Then, a preset meta-learning rate, which controls the overall update step size of the model parameters, is multiplied by the scaling factor and the gradient of the loss function to obtain the update amount of the trainable weight parameters of the machine learning model. The meta-learning rate is a small positive number, such as 0.001, used to ensure the stability of parameter updates. Finally, the trainable weight parameters of the machine learning model in step S1 are adjusted according to this update amount. When the overall difference is large, the absolute value of the scaling factor is close to one, and the adjustment step size of the model parameters is large; when the overall difference is small, the absolute value of the scaling factor is close to zero, and the adjustment step size of the model parameters is small. The effect of this adjustment loop is that it transforms the system-level performance monitoring signal (overall difference) into a direct guide for model parameter optimization, realizing closed-loop optimization from global system efficiency to micro-model parameters, enabling the model to continuously adapt to changes in network node behavior. The decision parameter adjustment targets the scale adjustment coefficient included in the dynamic node number determination formula in step S3. Specifically, R1: Based on the actual consensus time and the consistency of actual node responses, estimate an ideal number of verification nodes under the current balance between efficiency and security. For example, an empirical model can be established: when the actual response consistency is lower than a preset threshold (e.g., 95%), the number of nodes may be insufficient; when the consensus time is too long, the number of nodes may be excessive. The ideal number of nodes can be obtained through interpolation or a lookup table under this empirical model. R2: Calculate a direction discrimination value, which is equal to the dynamic verification node number determined in step S3. The actual size of the point set is subtracted from the ideal number of verification nodes, and then divided by the ideal number of verification nodes. A sign function is applied to this directional discriminant value to obtain a directional signal, which indicates whether the current node size is too large or too small. A sign function output of +1 indicates a large size, and the size adjustment coefficient should be reduced; an output of -1 indicates a small size, and the size adjustment coefficient should be increased. R3 calculates an adjustment criterion composed of both immediate and historical differences. The immediate difference component is the product of the overall difference degree of the current round and a preset immediate difference weighting factor. The immediate difference weighting factor is used to balance the influence of the current round's difference and historical differences; for example, it can be set to 0. 7 indicates a greater reliance on real-time information in the current round; the historical difference component is the comprehensive difference degree of several past rounds, multiplied by a decay weight with a preset historical error decay factor as the base and the time difference from the current round as the exponent, then summed, and finally multiplied by the difference between the number one and the real-time difference weight factor; the historical error decay factor is less than 1, for example 0.9, so that the impact of the difference degree in the past is smaller, realizing the "memory" of historical trends but also the "forgetting"; the real-time difference component and the historical difference component are added to obtain the adjustment basis; the introduction of the historical difference component (cumulative error term) can smooth the impact of single fluctuations and correct the systematic bias of the parameters. To prevent parameters from oscillating around their optimal values, R4 multiplies the direction signal, a preset adjustment strength coefficient for controlling the overall adjustment amplitude, and the adjustment basis to obtain the adjustment amount for the scale adjustment coefficient. The adjustment strength coefficient is a small positive number, such as 0.1, used to control the amplitude of each adjustment and avoid over-adjustment. R5 updates the scale adjustment coefficient in the dynamic node number determination formula in step S3 within the preset lower and upper limits of the scale adjustment coefficient. The lower and upper limits are used to ensure that the scale adjustment coefficient is within a reasonable range, for example, the lower limit is 1.0 and the upper limit is 10.0. This decision parameter adjustment loop continuously fine-tunes the key decision parameter (scale adjustment coefficient), enabling the resource scheduling strategy of the entire system (i.e., the determination of the size of the dynamic verification node set) to self-optimize based on long-term operating performance feedback, thereby continuously improving the system's ability to balance efficiency and safety.

[0024] Example 2 This embodiment provides, for example Figure 2 The blockchain communication system for secure sharing of bone tumor data shown includes: a node reputation and behavior prediction module, an institutional trust dynamic assessment module, a dynamic verification networking and decision-making module, and a real-time monitoring and feedback adjustment module, wherein; Node Reputation and Behavior Prediction Module: When a bone tumor data sharing transaction verification request is received, the module builds and updates a machine learning model based on the historical transaction verification records of the verification node to output the dynamic reputation score of each verification node and the expected behavior assessment result of the current transaction. The Institutional Trust Dynamic Assessment Module is used to construct a two-layer trust network based on dynamic reputation scores, attribute information of the institution to which the verification node belongs, and the current transaction context, and to calculate the contextual trust level of each institution in this transaction by combining an attention mechanism. Dynamic verification network and decision-making module: It is used to integrate the expected behavior assessment results and contextual trust level, select verification nodes from the verification nodes to form a dynamic verification node set and assign differentiated verification tasks. The size of the set is calculated by a dynamic node number determination formula containing an adjustable size adjustment coefficient. The input of the formula includes the network average trust level confidence obtained based on contextual trust level and the transaction risk score calculated based on the current transaction information. Real-time monitoring and feedback adjustment module: When performing verification on a dynamic set of verification nodes, it collects actual verification performance data and compares it with the expected behavior evaluation results, and adjusts the machine learning model and scale adjustment coefficients based on the differences.

[0025] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0026] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0027] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0028] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0029] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0030] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0031] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A blockchain communication method for secure sharing of bone tumor data, characterized in that, Specifically, the following steps are included: Step S1: In response to receiving a bone tumor data sharing transaction verification request in the blockchain network, construct and update a machine learning model for quantifying the reputation of verification nodes and predicting their verification behavior based on the historical transaction verification records of the verification nodes; the historical transaction verification records include the historical verification success rate, historical average response latency, and participation frequency distribution of each verification node in the verification of different subtypes of bone tumor data transactions; The machine learning model analyzes historical transaction verification records and outputs a dynamic reputation score for each verification node and an assessment of the expected behavior of the current transaction. Step S2: Based on the dynamic reputation score of the verification node, the attribute information of the institution to which the verification node belongs, and the specific contextual features of the current transaction, a dynamic two-layer trust network is generated. Institutional attribute information includes historical cooperation data between institutions; the physical node layer of the two-layer trust network corresponds to each verification node, and the logical institution layer corresponds to the institution to which each verification node belongs; using graph embedding technology and attention mechanism, combined with the specific contextual features of the current transaction, the contextual trust level of each institution in this transaction is calculated. Step S3: Based on the expected behavior assessment results of the verification nodes and the contextual trust level of the institution, execute optimization decisions. Optimization decisions include: selecting verification nodes from the verification nodes to form a dynamic verification node set based on the contextual trust level of the institution and the expected behavior assessment results of the verification nodes, and assigning verification tasks to the verification nodes in the dynamic verification node set. The size of the dynamic verification node set is calculated using a dynamic node number determination formula based on the network average trust confidence level calculated from the contextual trust level and the transaction risk score calculated by extracting the transaction amount and data sensitivity level from the current transaction information. The dynamic node number determination formula includes an adjustable size adjustment coefficient. Step S4: During the verification process of the dynamic verification node set, collect actual verification performance data in real time. The actual verification performance data includes consensus achievement time, node response consistency, and node response latency. Compare the actual verification performance data with the expected behavior evaluation results output by the machine learning model in step S1. Based on the differences generated by the comparison, execute a feedback adjustment process. The feedback adjustment process is used to adjust the machine learning model in step S1 and to adjust the scale adjustment coefficient included in the dynamic node number determination formula in step S3.

2. The blockchain communication method for secure sharing of bone tumor data according to claim 1, characterized in that: In step S1, based on the historical transaction verification records of the verification nodes, a machine learning model for quantifying the reputation of verification nodes and predicting their verification behavior is constructed and updated. This includes constructing time-series feature data as input to the machine learning model, specifically: From the distributed ledger of the blockchain, retrieve the historical transaction verification records of all verification nodes within a preset time window; for each verification node, construct a time-series feature vector sequence from its historical transaction verification records; for each historical time slice, statistically analyze the feature data of the verification node within that time slice to form a time-series feature vector; arrange these vectors in chronological order to form the time-series feature vector sequence of the verification node; each time-series feature vector in the time-series feature vector sequence corresponds to a historical time slice, and this time-series feature vector contains the following multiple dimensions of feature data statistically obtained by the verification node within that historical time slice; the multiple dimensions of feature data specifically include: A1. Historical verification success rate, which represents the proportion of transactions that were successfully verified by the verification node within the historical time slice to the total number of verified transactions. A2. Historical average response latency represents the average time taken for the verification node to respond to and complete verification for all verification requests within the historical time slice. A3, Participation Frequency Distribution Vector, represents the participation frequency distribution of the verification node in verifying different subtypes of bone tumor data sharing transactions within this historical time slice. The subtypes include image data transactions, genomic data transactions, and pathology report data transactions.

3. The blockchain communication method for secure sharing of bone tumor data according to claim 2, characterized in that: The machine learning model is a recurrent neural network model based on an encoder-decoder architecture, and the recurrent neural network model contains multiple trainable weight parameters. The machine learning model analyzes historical transaction verification records and outputs a dynamic reputation score for each verification node, along with an assessment of the expected behavior for the current transaction. Specifically, this includes: The temporal feature vector sequence is input into the encoder part of the recurrent neural network model. The encoder encodes the sequence and outputs a hidden state vector that represents the long-term behavior pattern of the verification node. The decoder part of the recurrent neural network model performs the following operations based on the hidden state vector and the type characteristics of the current bone tumor data sharing transaction to be verified: B1. Predict the behavior tendency of the verification node for the current transaction and output a behavior prediction label. This behavior prediction label constitutes the expected behavior evaluation result. The value of the behavior prediction label includes three categories: reliable, risky, and unstable. B2. Calculate the dynamic reputation score of the verification node through a reputation fusion calculation process. The specific process of this reputation fusion calculation is as follows: B2.

1. Perform a nonlinear transformation on the hidden state vector to obtain the basic reputation vector that represents the credibility of the historical behavior of the verification node itself. B2.2 Obtain the historical reputation scores of all neighboring verification nodes directly connected to the target verification node in the current reputation calculation in a preset verification node relationship network in the previous evaluation period, and determine the corresponding connection weights based on the historical interaction success rate between the target verification node and each neighboring verification node. Calculate a weighted average value based on the historical reputation scores and corresponding connection weights of all neighboring verification nodes as the network consensus reputation value. B2.

3. Perform a first linear transformation on the basic reputation vector to obtain the self-behavioral reputation component; multiply the network consensus reputation value by a preset network influence adjustment parameter to obtain the network influence reputation component; add the self-behavioral reputation component and the network influence reputation component to obtain the original reputation value. B2.4 Input the original reputation value into an S-shaped growth function for nonlinear mapping, restrict the mapping output to between zero and one, and use the result as the dynamic reputation score of the target verification node at the current time.

4. A blockchain communication method for secure sharing of bone tumor data according to claim 3, characterized in that: In step S2, a dynamic two-layer trust network comprising a physical node layer and a logical mechanism layer is constructed. The specific process is as follows: A dynamic two-layer trust network is constructed based on the affiliation between verification nodes and their affiliated institutions; The dynamic two-layer trust network consists of a physical node layer and a logical mechanism layer. Each physical node in the physical node layer corresponds to a verification node, and each physical node is assigned a node attribute. The value of the node attribute is the dynamic reputation score of the verification node corresponding to the physical node, obtained from step S1. Each logical organization node in the logical organization layer corresponds to an organization to which the verification node belongs. A connection edge is constructed for each pair of logical organization nodes in the logical organization layer, and an edge weight attribute is assigned to the connection edge. The value of the edge weight attribute is calculated based on the historical cooperation data between the organizations corresponding to the pair of logical organization nodes in the organization attribute information. Then, graph embedding learning is performed on the physical node layer and the logical mechanism layer respectively. Through graph embedding learning, each physical node in the physical node layer is mapped to a first embedding vector of fixed length, and each logical mechanism node in the logical mechanism layer is mapped to a second embedding vector of fixed length.

5. A blockchain communication method for secure sharing of bone tumor data according to claim 4, characterized in that: The process of using graph embedding technology and attention mechanisms, combined with the specific contextual features of the current transaction, to calculate the contextual trust level of each institution in this transaction is as follows: Extract the specific contextual features of the current bone tumor data sharing transaction to be verified. The specific contextual features include the transaction amount and the data sensitivity level. Based on these features, a transaction contextual feature vector is constructed. Next, using the second embedding vector of each logical mechanism node in the logical mechanism layer as the basic representation, and introducing the transaction context feature vector as the global context, a multi-head attention mechanism is used to dynamically calculate the association strength between any central logical mechanism node as the target and each of its neighboring logical mechanism nodes in the logical mechanism layer network under the current transaction context. The calculation process for the association strength is as follows: The query representation is obtained by multiplying the second embedding vector of the central logical mechanism node by a first learnable parameter matrix; the key representation is obtained by multiplying the second embedding vector of the neighboring logical mechanism nodes by a second learnable parameter matrix; the dimensions of the first and second learnable parameter matrices are determined by the length of the second embedding vector and a preset attention space dimension; then the scaled dot product of the query representation and the key representation is calculated, and the calculation results of all neighboring logical mechanism nodes are normalized exponentially to obtain the dynamic influence weight of the neighboring logical mechanism node on the central logical mechanism node; Based on this, the contextual trust level of the institution corresponding to the central logical institution node for the current transaction is calculated; the calculation process for this contextual trust level is as follows: C1. For each neighboring logical institution node of the central logical institution node, fuse and encode the second embedding vector of the neighboring logical institution node with the historical cooperation data represented by the edge weight attribute of the edge connecting the two logical institution nodes to generate a neighbor contribution feature. Then multiply this neighbor contribution feature by the dynamic influence weight corresponding to the neighboring logical institution node to obtain its weighted contribution. Finally, sum the weighted contributions of all neighboring logical institution nodes to obtain the institution network trust component. C2. From the output of step S1, obtain the dynamic reputation scores of all verification nodes belonging to the central logical mechanism node, calculate the arithmetic mean of these dynamic reputation scores, and obtain the reputation components of the subordinate nodes. C3. Multiply the subordinate node reputation component by a preset institutional node trust transfer coefficient with a value range between zero and one, and then add it to the institutional network trust component to obtain an original trust value. C4. Input the original trust value into an S-shaped growth function for non-linear mapping, and restrict the mapping output value to between zero and one. The final output value is the contextual trust value of the institution corresponding to the central logical institution node in this transaction.

6. A blockchain communication method for secure sharing of bone tumor data according to claim 5, characterized in that: In step S3, based on the network average trust confidence level calculated from contextualized trust level and the transaction risk score extracted from the current transaction information and the data sensitivity level, the size of the dynamic verification node set is calculated using a dynamic node number determination formula, specifically including: First, calculate the network average confidence level. The calculation process is as follows: Obtain the contextualized trust level of all institutions in this transaction calculated in step S2, and denote the total number of institutions as the first total. Calculate the arithmetic mean of these contextualized trust levels, and denote it as the first mean. Calculate the standard deviation of these contextualized trust levels relative to the first mean, and denote it as the first standard deviation. Then, for each institution, calculate the fourth power of the difference between its contextualized trust level and the first mean, and obtain the fourth power deviation value for that institution. Calculate the product of the first total and the fourth power of the first standard deviation, and denote it as the first denominator. Calculate the sum of the fourth power deviation values ​​of all institutions, divide the sum by the first total, and obtain the mean of the fourth power deviation values. Then, subtract the ratio of the mean of the fourth power deviation values ​​to the fourth power of the first standard deviation from the first value. The result is the network average confidence level. Secondly, the transaction risk score is calculated. The calculation process is as follows: extract the transaction amount and data sensitivity level from the current transaction information; add one to the transaction amount, take the natural logarithm, and then multiply by a first weighting coefficient to obtain the amount risk component; square the data sensitivity level and then multiply by a second weighting coefficient to obtain the sensitivity risk component; add the amount risk component and the sensitivity risk component, input the sum into a hyperbolic tangent function for mapping, and restrict the mapping result to between zero and one. This result is the transaction risk score. Finally, the average network trust level and transaction risk score are substituted into the dynamic node number determination formula. The calculation process of this dynamic node number determination formula is as follows: E1. Calculate the difference between the number one and the network average confidence level. E2. Calculate the product of the transaction risk score and a preset risk sensitivity amplification coefficient, and then use the natural constant e as the base and this product as the exponent to calculate the exponential function value. E3. Multiply a preset scale adjustment coefficient, the difference calculated by E1, and the exponential function value calculated by E2 to obtain a scale adjustment amount. E4. Add a preset minimum number of safe nodes to the size adjustment amount to obtain the size calculation value; E5. Round the calculated size up to the nearest integer, which is the size of the dynamic verification node set.

7. A blockchain communication method for secure sharing of bone tumor data according to claim 6, characterized in that: Based on the institution's contextualized trust level and the expected behavior assessment results of the verification nodes, verification nodes are selected from the verification nodes to form a dynamic verification node set, and verification tasks are assigned to the verification nodes in the dynamic verification node set, specifically including: Based on the size of the dynamic verification node set, perform node filtering and task assignment. First, the quota allocation for each institution is carried out. Specifically, for each institution, its contextual trust level is multiplied by the size of the dynamic verification node set, and then divided by the sum of the contextual trust levels of all institutions. The result is rounded up, and the rounded value is the initial quota number of nodes obtained by the institution. Then, within each institution, a two-step screening method is used to select the best nodes within the institution; Ultimately, the verification nodes selected by all institutions using the two-step screening method together constitute a dynamic set of verification nodes; When assigning verification tasks to verification nodes in the dynamic verification node set, a differentiated assignment strategy is adopted: for all verification nodes from the same institution that are selected into the dynamic verification node set, the verification node with the highest dynamic reputation score is selected and assigned a complete verification task; sampling verification tasks are assigned to other verification nodes within the same institution that are selected into the dynamic verification node set.

8. A blockchain communication method for secure sharing of bone tumor data according to claim 7, characterized in that: In step S4, the specific process of collecting actual verification performance data in real time and comparing the actual verification performance data with the expected behavior evaluation results output by the machine learning model in step S1 is as follows: During the verification process of the current bone tumor data sharing transaction executed by the dynamic verification node set, the actual performance data of this verification task is collected in real time through deployed monitoring probes, forming an actual performance data triplet; the actual performance data triplet consists of the following three indicators: U1, Actual consensus time, is the length of time from the start of verification at the first verification node to the completion of consensus. U2. Actual node response consistency, which is the proportion of the number of nodes that output the same correct verification result in the set of dynamic verification nodes to the total number of nodes in the set. U3, Average Response Time of Actual Nodes, is the average time taken by all verification nodes in the dynamic verification node set from receiving the verification task to submitting the verification result. Simultaneously, data relevant to this verification task is extracted from the expected behavior evaluation results and dynamic reputation scores output by the machine learning model in step S1, serving as expected performance data to form expected performance data triplets; the expected performance data triplets consist of the following three estimation metrics: P1, Expected consensus time, is estimated based on the predicted behavior labels of each verification node selected into the dynamic verification node set and its historical average response delay. P2. Expected node response consistency is estimated based on the proportion of the number of verification nodes whose predicted behavior label is marked as "reliable" among the verification nodes selected into the dynamic verification node set to the total number of nodes in the set. P3, Expected average node response latency, is estimated based on the historical average response latency of each verification node selected in the dynamic verification node set; Next, each indicator in the actual performance data triplet is compared with the corresponding estimated indicator in the expected performance data triplet, and the relative deviation between each pair of indicators is calculated. Then, a dynamic weight is assigned to the relative deviation of each type of performance metric; Finally, the relative deviation of each performance index is multiplied by a preset deviation sensitivity coefficient, and then input into an S-shaped growth function for nonlinear compression mapping. The mapping result is multiplied by the corresponding dynamic weight, and the results of all three performance indices after weighted processing are summed to obtain a comprehensive difference degree.

9. A blockchain communication method for secure sharing of bone tumor data according to claim 8, characterized in that: Based on the differences generated by the comparison, a feedback adjustment process is executed, specifically including: Model parameter tuning is performed on the machine learning model in step S1, specifically as follows: First, the comprehensive difference is multiplied by a preset model adjustment sensitivity coefficient, and then the multiplication result is input into a hyperbolic tangent function for mapping to obtain a value between negative one and positive one, which is used as a scaling factor for updating model parameters. Secondly, using a recent batch of data containing actual validation performance labels, the gradient of the current trainable weight parameters of the machine learning model with respect to an adaptive loss function is calculated. Then, a preset meta-learning rate, which controls the overall update step size of the model parameters, is multiplied by the scaling factor and the gradient of the loss function to obtain the update amount of the trainable weight parameters of the machine learning model. Finally, the trainable weight parameters of the machine learning model in step S1 are adjusted based on this update amount. The adjustment of the decision parameters is applied to the scale adjustment coefficient included in the formula for determining the number of dynamic nodes in step S3, specifically as follows: R1. Based on the consistency between the actual consensus time and the actual node response, estimate an ideal number of verification nodes under the current balance between efficiency and security. R2. Calculate a direction discrimination value, which is equal to the actual size of the dynamic verification node set determined in step S3 minus the ideal number of verification nodes, and then divided by the ideal number of verification nodes; take the sign function of the direction discrimination value to obtain a direction signal; R3. Calculate an adjustment basis consisting of immediate difference and historical difference; the immediate difference component is the product of the comprehensive difference degree of the current round and a preset immediate difference weighting factor; the historical difference component is the comprehensive difference degree of the past several rounds, which is multiplied by a decay weight with a preset historical error decay factor as the base and the time difference from the current round as the exponent, then summed, and finally multiplied by the difference between the number one and the immediate difference weighting factor; add the immediate difference component and the historical difference component to obtain the adjustment basis; R4. Multiply the direction signal, a preset adjustment intensity coefficient for controlling the overall amplitude, and the adjustment basis to obtain the adjustment amount of the scale adjustment coefficient. R5. Based on this adjustment amount, within the preset lower and upper limits of the scale adjustment coefficient, update the scale adjustment coefficient in the dynamic node number determination formula described in step S3.

10. A blockchain communication system for secure sharing of bone tumor data, applied to the blockchain communication method for secure sharing of bone tumor data as described in any one of claims 1-9, characterized in that: Specifically, it includes: The module includes a node reputation and behavior prediction module, an institutional trust dynamic assessment module, a dynamic verification networking and decision-making module, and a real-time monitoring and feedback adjustment module. Node Reputation and Behavior Prediction Module: When a bone tumor data sharing transaction verification request is received, the module builds and updates a machine learning model based on the historical transaction verification records of the verification node to output the dynamic reputation score of each verification node and the expected behavior assessment result of the current transaction. Institutional Trust Dynamic Assessment Module: This module is used to construct a two-layer trust network based on the dynamic reputation score, the attribute information of the institution to which the verification node belongs, and the current transaction context, and to calculate the contextual trust level of each institution in this transaction by combining an attention mechanism. Dynamic verification network and decision-making module: It is used to integrate the expected behavior evaluation results and contextual trust level, select verification nodes from the verification nodes to form a dynamic verification node set and assign differentiated verification tasks. The size of the set is calculated by a dynamic node number determination formula containing an adjustable size adjustment coefficient. The input of the formula includes the network average trust level confidence obtained based on the contextual trust level and the transaction risk score calculated based on the current transaction information. Real-time monitoring and feedback adjustment module: When the dynamic verification node set performs verification, it collects the actual verification performance data and compares it with the expected behavior evaluation results, and adjusts the machine learning model and the scale adjustment coefficient according to the difference feedback.

Images