Blockchain-based methods and systems for storing medical research data

Abstract

This disclosure provides a blockchain-based method and system for storing medical research data. The method includes: a data platform, upon receiving target medical research data, sending the data to a data storage system; the data storage system obtaining data feature vectors based on the data structure type; the system fusing the data feature vectors to obtain target data features, and comparing these features with the data features of previously stored medical research data; and the system determining storage information and storing it on the blockchain when the similarity between the target data features and the data features of all previously stored medical research data is less than or equal to a preset level. This disclosure addresses the risks of tampering and infringement of medical research data by storing its data features on the blockchain, preventing data tampering, and achieving data storage for medical research.

Description

Technical Field

[0001] This disclosure relates to the field of medical research data storage technology, and in particular to a blockchain-based method and system for medical research data storage. Background Technology

[0002] With the development of medical research, medical research projects are of great significance to promoting medical progress. Medical research projects can refer to an academic or practical process of exploring and verifying specific medical problems or goals through systematic methods. The essence of medical research projects is to use scientific means to solve practical medical problems and promote innovation in medical or healthcare knowledge.

[0003] For medical research data, the arrival of the digital age has brought convenience to information sharing, but it has also posed new challenges to the authenticity and security of data. When research data is shared online, there is a risk of it being tampered with and infringed upon, which reduces the credibility of research results and fails to protect the rights and interests of researchers. Summary of the Invention

[0004] This disclosure provides a blockchain-based method and system for storing medical research data, aiming to at least address the risks of data tampering and infringement in related technologies, and the inability to ensure the authenticity and reliability of medical research results. The technical solution of this disclosure is as follows:

[0005] According to a first aspect of this disclosure, a blockchain-based method for storing medical research data is provided. This method is applied to a medical research data storage system, which includes a data platform, a data storage system, and a local database. The method comprises: the data platform, in response to receiving target medical research data uploaded by a first terminal, sending the target medical research data to the data storage system; the data storage system encoding the target medical research data according to data structure types to obtain data feature vectors corresponding to data of each data structure type in the target medical research data; the data storage system fusing the data feature vectors to obtain target data features, and comparing the target data features with data features of medical research data already stored in the local database; and the data storage system, in response to the comparison result indicating that the similarity between the target data features and the data features of all stored medical research data is less than or equal to a preset level, determining storage information corresponding to the target medical research data and storing the storage information in the blockchain, wherein the storage information includes information about the target data features and the upload time of the target medical research data.

[0006] Optionally, the project evidence storage system obtains the target data features by: fusing data feature vectors corresponding to different data structure types at least once to obtain cross feature vectors, wherein the preset data structure types include structured data, semi-structured data, and unstructured data; determining the information contribution of each data feature vector and / or each cross feature vector in the target medical project data, wherein the information contribution represents the importance of the data feature vector in expressing the information of the target medical project data; and determining the target data features based on the data feature vectors, the cross feature vectors, and the information contribution.

[0007] Optionally, determining the target data feature based on the data feature vector, the cross feature vector, and the information contribution degree includes: determining the correlation information between each data feature vector and the medical topic data type according to a preset medical topic data type; filtering target feature vectors from the data feature vectors based on the correlation information, wherein the target feature vector is a data feature vector among the data feature vectors whose correlation with the medical topic data type is higher than a preset level; and determining the target data feature based on the target feature vector, the cross feature vector, and the information contribution degree corresponding to the target feature vector and / or the cross feature vector.

[0008] Optionally, the target data feature is a gene value, wherein determining the target data feature based on the target feature vector, the cross feature vector, and the information contribution of the target feature vector includes: inputting the target feature vector, the cross feature vector, and the information contribution of the target feature vector into a pre-trained feature calculation model, and outputting the target data feature from the feature calculation model, wherein the feature calculation model includes a residual network, an attention module, and a prediction module, and is trained by: obtaining a training sample set of medical research data, wherein the training sample set includes multiple training samples and sample labels for each training sample, wherein each training sample includes multiple sample targets. The feature calculation model comprises a feature vector, a sample cross feature vector, and the information contribution of each sample target feature vector, wherein the sample label represents the gene value of the training sample; the semantic information of each sample target feature vector of each training sample is extracted using the residual network; the target feature vectors of each training sample are input into the attention module to obtain the attention fusion feature vector corresponding to each training sample; the gene value of each training sample is predicted using the prediction module based on the semantic information of each sample target feature vector, the attention fusion feature vector, the sample cross feature vector, and the information contribution, to obtain the prediction result; and the feature calculation model is trained based on the prediction result and the sample label to obtain the trained feature calculation model.

[0009] Optionally, the data storage method further includes: in response to receiving a query request from a second terminal for the target medical research data, the data platform sends the queried target medical research data to the second terminal; in response to receiving a reference request from the second terminal for the queried target medical research data, the data platform queries the blockchain for a first feature public key of the target medical research data stored on the blockchain, and compares the first feature public key with a second feature public key of the target medical research data stored on the data platform, wherein the first feature public key and the second feature public key indicate the identity information of the uploader of the target medical research data; in response to the first feature public key and the second feature public key being consistent, the data platform verifies the signature information of the target medical research data; in response to the first feature public key and the second feature public key being inconsistent, the data platform provides the second terminal with a warning message indicating that the target medical research data is at risk of being tampered with.

[0010] Optionally, verifying the signature information of the target medical research data includes: the data platform querying the blockchain for first signature information of the target medical research data stored on the blockchain, and comparing the first signature information with second signature information of the target medical research data stored on the data platform; in response to the first signature information and the second signature information being consistent, the data platform verifying the second signature information using the second feature public key; in response to the successful verification of the second signature information, the data platform sending the target medical research data to the second terminal, wherein the first signature information and the second signature information are generated based on a feature private key corresponding to the feature public key of the target medical research data.

[0011] Optionally, the data storage method further includes: in response to receiving a query request for the target medical topic data from a second terminal, before executing the step of sending the queried target medical topic data to the second terminal, the data platform queries the blockchain for the first feature public key and compares the first feature public key with the second feature public key; in response to the first feature public key and the second feature public key being consistent, the data platform verifies the signature information of the target medical topic data; in response to the first feature public key and the second feature public key being inconsistent, the data platform provides the second terminal with a warning message indicating that the target medical topic data is at risk of being tampered with.

[0012] According to a second aspect of this disclosure, a medical research project data storage system is provided. The system includes a data platform, a project storage system, and a local database. Upon receiving target medical research project data uploaded by a first terminal, the data platform sends the target medical research project data to the project storage system. The project storage system determines target data features of the target medical research project data and compares these features with data features of medical research project data already stored in the local database. Upon receiving a comparison result indicating that the similarity between the target data features and the data features of all stored medical research project data is less than a preset level, the project storage system determines storage information corresponding to the target medical research project data and stores the storage information on a blockchain. The storage information includes information about the target data features and the upload time of the target medical research project data.

[0013] According to a third aspect of this disclosure, a computer-readable storage medium is provided that, when instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform a blockchain-based method for storing medical research data according to this disclosure.

[0014] According to a fourth aspect of this disclosure, a computer program product is provided, including computer-executable instructions that, when executed by at least one processor, implement the blockchain-based medical subject data storage method according to this disclosure.

[0015] The technical solution provided in this disclosure brings at least the following beneficial effects:

[0016] According to the medical research data preservation scheme disclosed herein, the research data preservation system can encode the research data according to the data structure type and obtain corresponding data feature vectors. These feature vectors can then be fused to obtain target data features. The data features of the currently uploaded medical research data can be compared with those of previously stored medical research data. If the similarity is below a preset level, the preservation information of the currently uploaded medical research data can be stored on the blockchain. This preservation information includes the data features of the medical research data and the upload time. This minimizes the possibility of the uploaded medical research data infringing on existing medical research data and stores the data features on the blockchain to prevent tampering, thus achieving the preservation of medical research data.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0019] Figure 1 This is a schematic block diagram of a medical subject data storage system according to exemplary embodiments of the present disclosure.

[0020] Figure 2 This is a schematic flowchart of a method for storing medical subject data according to exemplary embodiments of the present disclosure.

[0021] Figure 3 This is a schematic diagram illustrating an example of storing medical subject data in a medical subject data storage method according to an exemplary embodiment of the present disclosure.

[0022] Figure 4 This is an illustrative flowchart illustrating an example of determining target data characteristics in a medical subject data preservation method according to an exemplary embodiment of the present disclosure.

[0023] Figure 5This is an illustrative flowchart illustrating an example of determining target data features based on the filtering of feature vectors in a medical subject data storage method according to exemplary embodiments of the present disclosure.

[0024] Figure 6 This is a schematic flowchart illustrating an example of training a feature calculation model in a medical subject data storage method according to an exemplary embodiment of the present disclosure.

[0025] Figure 7 This is a schematic diagram of the feature calculation model in a medical subject data storage method according to an exemplary embodiment of the present disclosure.

[0026] Figure 8 This is a schematic diagram illustrating an example of referencing medical subject data in a medical subject data storage method according to an exemplary embodiment of the present disclosure.

[0027] Figure 9 This is a schematic diagram illustrating an example of medical issue data rights protection in a medical issue data preservation method according to an exemplary embodiment of the present disclosure. Detailed Implementation

[0028] In order to enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0029] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0030] It should be noted that the phrase "at least one of several items" in this disclosure refers to three parallel cases: "any one of the several items", "a combination of any number of the several items", and "all of the several items". For example, "including at least one of A and B" includes the following three parallel cases: (1) including A; (2) including B; (3) including A and B. As another example, "performing at least one of step one and step two" indicates the following three parallel cases: (1) performing step one; (2) performing step two; (3) performing both step one and step two.

[0031] As mentioned earlier, in related technologies, there are risks of medical research data being tampered with and infringed upon, and the authenticity and reliability of research results cannot be guaranteed.

[0032] In view of the above problems, the exemplary embodiments of this disclosure propose a blockchain-based method for storing medical research data, a system for storing medical research data, a computer-readable storage medium, and a computer program product, which can solve or at least alleviate the above problems.

[0033] In a first aspect of an exemplary embodiment of this disclosure, a blockchain-based method for storing medical research data is provided.

[0034] The blockchain-based medical research data storage method according to the exemplary embodiments of this disclosure can be applied to scenarios where a user terminal (e.g., a first terminal) interacts with a medical research data storage system. The medical research data storage system may include a data platform, a data storage system, and a local database. A user can input target medical research data to be stored on the user terminal and send the target medical research data to the data platform. Upon receiving the target medical research data uploaded by the user terminal, the data platform sends the target medical research data to the data storage system. The data storage system encodes the target medical research data according to the data structure type, obtaining data feature vectors corresponding to the data of each data structure type in the target medical research data. The data storage system fuses the data feature vectors to obtain target data features of the target medical research data and compares these target data features with the data features of medical research data already stored in the local database. Upon receiving a comparison result indicating that the similarity between the target data features and the data features of all stored medical research data is less than a preset level, the data storage system determines the storage information corresponding to the target medical research data and stores the storage information on the blockchain. The storage information includes information about the target data features and the upload time of the target medical research data.

[0035] Here, the data platform, project documentation system, and local database can be deployed on a server. User terminals and the server can communicate bidirectionally to transmit data. The aforementioned user terminals can be devices such as tablets, laptops, digital assistants, and wearable devices. The server can be a standalone server, a server cluster, a cloud computing platform, or a virtualization center.

[0036] The blockchain-based method for storing medical research data according to exemplary embodiments of this disclosure can minimize the possibility of uploaded medical research data infringing on existing medical research data, and can store the data characteristics of medical research data on the blockchain to prevent the medical research data from being tampered with, thereby achieving the storage of medical research data.

[0037] The following will refer to Figures 1 to 9 Examples of a blockchain-based method for storing medical research data according to embodiments of this disclosure are described.

[0038] This method for preserving medical research data can be applied to a medical research data preservation system. Before describing this method, we will first refer to... Figure 1 Describe the data storage system for medical research projects.

[0039] like Figure 1 As shown, the medical research project data storage system 100 may include a data platform 110, a research project storage system 120, and a local database 130. The data platform 110 can receive medical research project data uploaded by user terminals 200 and send the data to the research project storage system 120. The research project storage system 120 can determine the data characteristics of the medical research project data and compare the data characteristics between different medical research project data to determine the similarity between them, thereby assessing whether the medical research project data poses a risk of infringement. The local database 130 can store the medical research project data and related information. Both the data platform 110 and the research project storage system 120 are bidirectionally connected to the local database 130.

[0040] In addition, the data platform 110 and the project evidence storage system 120 can also be bidirectionally connected to the blockchain system 300 to store data on the blockchain or retrieve stored data from the blockchain.

[0041] based on Figure 1 Medical research data storage system, such as Figure 2 As shown, the data preservation method for this medical research project may include the following steps:

[0042] In step S210, in response to receiving the target medical research data uploaded by the first terminal, the data platform sends the target medical research data to the research data storage system.

[0043] Here, the target medical research data can be research data from academic research projects, such as experimental data. The first terminal can be a user terminal, such as... Figure 3 As shown, researchers can log in to the data platform, enter project information, and upload original target medical project data to request its publication on the platform. Before publishing the target medical project data, the data platform can send the data to a project registration system for data feature comparison to avoid the risk of infringement between the current medical project data and already published medical project data.

[0044] Furthermore, as an example, in response to receiving the target medical research data uploaded by the first terminal, before executing the step of sending the target medical research data to the research data storage system, the data platform can also perform secondary identity verification on the user who is logged in and uploaded the target medical research data. For example, the secondary identity verification may include verifying the user information entered by the user during registration. The process of entering and storing user information will be described in detail later.

[0045] Specifically, users such as researchers can submit medical research data within the specified timeframe. After logging into the data platform, they can add or find their research projects and enter the raw data. Once the final version of the data is entered within the specified time, the researchers can confirm and publish the medical research data. When publishing the data, the data platform can perform secondary identity verification on the user through comparison of user information such as fingerprints or iris scans.

[0046] Furthermore, after successful secondary identity verification, the data platform can associate the published medical research data with user information and blockchain transaction information related to that user information. After association, it can extract data feature information from the target medical research data, generate a feature private key from this data feature information, generate a corresponding feature public key using an elliptic curve algorithm, and generate signature information for the target medical research data using the user information's private key. The generated feature private key, feature public key, and signature information are then stored together in a local database (DB). After successful database storage, the data platform can store the data's feature public key and signature information in the blockchain system. The blockchain system returns transaction information (e.g., transaction ID), updating the corresponding record in the stored local database. The user information and its blockchain transaction information will be described in detail later.

[0047] Furthermore, in this paper, the process of generating signature information using a private key can employ cryptographic algorithms such as asymmetric encryption (RSA) and elliptic curve digital signature (ECDSA). In this way, the public key and signature information can be used to verify the signature using the corresponding verification algorithm. If the verification passes, it can be concluded that the message was indeed signed by the party holding the corresponding private key, and that the message has not been tampered with during transmission.

[0048] In step S220, the project evidence storage system encodes the target medical project data according to the data structure type, and obtains the data feature vector corresponding to the data of each data structure type in the target medical project data.

[0049] Here, the data structure type can be preset. For example, the data structure type can include structured data, semi-structured data, and / or unstructured data. Specifically, due to the special nature of medical research data, such as medical papers and reports, medical research data may include: experimental numerical values ​​or clinical data with structured representation; plain text descriptions with unstructured representation; and hierarchical structures that divide these structured and unstructured data into chapters or sections, which are semi-structured data.

[0050] In this step, data belonging to different data types in the target medical research data can be encoded separately, and different encoding methods can be used for different data types, so as to better reflect the core information to be conveyed by each data type.

[0051] In one example, the data structure type can include structured data and unstructured data, and unstructured data can include text and / or images.

[0052] Structured data can be encoded based on unique identifiers (such as identity identifiers (IDs) or hash values) or key fields (such as names or dates) to obtain data feature vectors. In this example, the unique identifiers and key fields of the structured data from two research topics can be compared, and similarity can be calculated, such as calculating the proportion of matching identifiers and fields.

[0053] Text in unstructured data can have its feature vectors obtained by extracting keywords, TF-IDF vectors, or N-gram models. In this example, the feature vectors of text from two different research topics can be compared by calculating the cosine similarity between them.

[0054] Images in unstructured data can have their data feature vectors obtained by extracting color histograms, texture features, or deep learning embedding vectors (such as ResNet). In this example, similarity can be compared by calculating the Euclidean distance between the data feature vectors of images from two different research topics.

[0055] In another example, the data structure type can include structured data, semi-structured data, and unstructured data.

[0056] As an example, for structured data, normalization can be performed to encode the structured data. This simplifies subsequent data processing or computation while maintaining the original data distribution. For example, the normalization process for structured data can be represented by the following equation (1):

[0057]

[0058] in, x represents a normalized structured data vector. struct Min(x) represents the structured data vector before normalization. struct ) and max(x struct These are the minimum and maximum values ​​in the dimension of the structured data, respectively. This formula can normalize structured data, such as experimental numerical values, to the [0,1] interval, facilitating subsequent fusion and calculations with other data.

[0059] Although the above example of normalization processing describes the encoding method for structured data, it is not limited to this. Other methods can also be used for encoding, such as Z-Score encoding.

[0060] As an example, for semi-structured data, encoding can be achieved by establishing recursive expressions between the current node and its parent and child nodes. Semi-structured data can be tree-like, such as the chapter hierarchy of a medical research report, and can be in XML / JSON format. Assume the feature vector of node v in the tree is e. v The parent node is p(v), and the set of child nodes is C(v). For example, a feature vector is a structured representation of a node in semi-structured data, used to capture the semantic and structural information of that node. As an example, in a tree structure, the parent node p(v) is the node directly above the current node; each node (except the root node) has one and only one parent node. The child nodes C(v) are the set of nodes directly below the current node; a node can contain zero or more child nodes. The encoding vector h of any node v... v For example, it can be calculated recursively using the following formula (2):

[0061]

[0062] Among them, h v E represents the encoding vector of node v. v h represents the feature vector of node v. u h represents the final encoded vector of the current node. p(v) This represents the final encoded vector of the parent node. Here, during the recursive encoding of semi-structured data, a hierarchical feature representation can be formed. Specifically, if it is necessary to analyze the semantics of a specific node in the tree (e.g., "whether there is plagiarism in the discussion chapter"), the encoded vector h of node v can be used. v The collection of encoded vectors from multiple nodes can form a global vector. If you need to compare the overall structural similarity of two documents (e.g., chapter order, hierarchical depth), you can use global vectors. In equation (2) above, σ is the activation function, for example, it can be the ReLU function σ(x) = max(0,x), Wself W child and W parent It is a learnable weight matrix. In this way, the tree structure information of semi-structured data can be transformed into a vector representation.

[0063] In equation (2) above, the encoding vector h v This represents a vector obtained after performing specific encoding operations on data of different data structure types. When processing semi-structured data, a vector obtained by establishing a recursive expression between the current node and its parent and child nodes is a type of encoded vector, reflecting the transformation of the data's structure and semantic information. Encoded vector h v It is mainly used to transform data with different structures into a processable vector form, preserving the structural and semantic information of the data, and preparing for the subsequent generation of more representative feature vectors. When analyzing the tree structure of semi-structured data, the encoded vector can help capture the semantic and structural information of the nodes.

[0064] Furthermore, encoding semi-structured data may involve complex computational processes, such as generating encoded vectors by establishing relationships between the current node and its parent and child nodes through recursive expressions. In the example above, the encoded vector h... v This represents a temporary encoding vector for the current node at an intermediate stage in the recursive computation process. It contains partial information about the current node and its surrounding structure (e.g., parent node or both parent and child nodes), but the final integration and transformation have not yet been completed. The encoding vector h of the current node... u As the final encoding vector, it is obtained after comprehensively considering the node's position in the entire tree structure, its relationship with other nodes, and related calculation rules. It more comprehensively and accurately reflects the node's features and semantic information in the entire semi-structured data. It is the result of a series of calculations and integrations, and is used for tasks such as more in-depth data feature analysis and comparison.

[0065] eigenvector e vFeature vectors can be a quantitative representation of the key content (or representative content or content characteristics) of medical research data, reflecting the core features of the data. Different data structure types (structured, semi-structured, unstructured) can each have corresponding feature vectors. For example, existing feature extraction methods can be used to extract corresponding feature vectors from data of various structure types. For instance, after encoding data of different structure types (structured, semi-structured, unstructured), further fusion and filtering operations can yield vectors representing medical research data. Similarly, target data features obtained after steps such as data feature vector fusion and determining information contribution can be feature vectors. Such feature vectors can be used to compare the similarity between medical research data and determine whether there is a risk of data infringement. By calculating the distance or similarity between the feature vectors of different medical research data, it can be determined whether newly uploaded data is similar to already stored data, thereby deciding whether to store the data as evidence or indicate the risk of infringement.

[0066] Here, as an example, for semi-structured data, different encoding methods can be used to transform the data in medical research data into feature vectors. Subsequently, the feature vectors obtained after encoding different data structure types can be fused to obtain cross-feature vectors. Methods such as concatenation and weighted summation can be used. Combining the cross-feature vectors and the information contribution of each vector, the final feature vector of the medical report is determined. The cross-feature vectors can be weighted and combined with their information contribution to obtain the final feature vector.

[0067] Furthermore, in equation (2) above, the global vector This vector represents a set of encoded vectors from multiple nodes, encompassing the structural and semantic features of various parts of a document containing medical research data. It can reflect the overall structural characteristics of the document from a macroscopic perspective. As an example, in some cases, comparing the global vectors of different documents... The similarity can also be used to assess the structural similarity of data from different medical topics and to determine whether there are potential associations or infringement risks.

[0068] As an example, unstructured data can be encoded through semantic vector generation. Here, semantic vector generation can be implemented, for example, using a Bidirectional Encoder Representations from Transformers (BERT) model. For instance, unstructured text data T can be input, and after semantic generation processing using a model like BERT, the semantic vector for each word in the text can be obtained. Here, i represents the position of the word. Assuming there are n words in the text, the semantic vector of the entire text... It can be obtained through average pooling. For example, it can be expressed by the following equation (3):

[0069]

[0070] in, A semantic vector representing unstructured data.

[0071] It should be noted that although examples of encoding structured data, semi-structured data, and unstructured data have been described above, the embodiments of this disclosure are not limited thereto. Other existing encoding methods can also be used to encode these data to obtain the corresponding data feature vectors.

[0072] In step S230, the project evidence storage system obtains the target data features of the target medical project data by fusing the data feature vectors, and compares the target data features with the data features of the medical project data already stored in the local database.

[0073] Specifically, medical research data that has been successfully published on the data platform is stored in a local database. Here, the local database refers to the database of the medical research data storage system. For example, the local database can be deployed on the same server or server cluster as the data platform and the research data storage system.

[0074] The research project evidence storage system can first determine the target data characteristics of the currently received medical research project data, and can also retrieve the data characteristics of previously stored medical research project data for comparison with the target data characteristics. Here, because the system determines the data characteristics of the medical research project data to be published each time it is published, these data characteristics are stored in a local database corresponding to the published data, for comparison with the data characteristics of subsequently uploaded medical research project data. The system provides functions such as evidence retrieval, evidence storage, and data characteristic calculation.

[0075] Data features can be a quantitative representation of medical research data, reflecting its key content. Since the research findings of different medical research projects are unlikely to be completely identical—for example, data features can possess specificity and uniqueness—comparing the similarity of data features allows for comparison of the content of the original medical research data, thereby determining whether issues such as infringement exist. As an example, data features could be gene values ​​from medical research data.

[0076] In this step, the project evidence storage system can compare the target data features with the stored data features by calculating the similarity between the target data features and the stored data features.

[0077] As an example, such as Figure 4 As shown, the project evidence storage system can obtain the target data features in the following ways:

[0078] In step S410, at least one fusion can be performed between data feature vectors corresponding to different data structure types to obtain cross feature vectors.

[0079] For example, the normalized structured data vector can be obtained through the method described in step S220 above. Semi-structured data vector after tree structure encoding and unstructured data semantic vectors In this step, the normalized structured data vector can be... Semi-structured data vector after tree structure encoding and unstructured data semantic vectors Concatenate them sequentially to form the initial fusion vector. In this context, ";" indicates a vector concatenation operation.

[0080] In one example, the initial fusion vector can be used as the aforementioned cross feature vector.

[0081] In another example, the data feature vectors of different data structure types in the initial fusion vector can be further fused to obtain the final cross feature vector.

[0082] In this example, the methods for fusing data feature vectors of different data structure types can include: feature cross-multiplication; and / or, bitwise addition.

[0083] Specifically, feature cross-multiplication can refer to cross-multiplying the feature values ​​of feature vectors of different data structure types to generate new feature values, which are then combined into cross feature vectors.

[0084] For example, we can set a feature value x in the feature vector of structured data. s (For example, the mean of experimental parameters), the word frequency of the preset keywords in the feature vector of unstructured data is x. n The new eigenvalue x generated by cross-multiplication new For: x new =x s ×x n As an example, the preset keywords can be words related to the feature values ​​in the structured data feature vector. For instance, based on the semi-structured data feature vector, the structured data feature vectors (e.g., x) under the same node can be used to select the feature values. s ) and unstructured data feature vectors (e.g., x nBy performing cross-multiplication, a relationship can be established between the textual descriptions in the medical research data and the experimental parameters / data, thereby better representing the research content in the medical research data.

[0085] Bitwise addition can refer to adding the feature values ​​of feature vectors of different data structure types bitwise to generate cross feature vectors.

[0086] For example, two feature vectors of different types of data can be represented as follows: and Generate a new feature vector by adding the positions. c i =a i +b i Where i = 1, 2, ..., m. As an example, as described above, for instance, structured data feature vectors (e.g., x) under the same node can be selected based on semi-structured data feature vectors. s ) and unstructured data feature vectors (e.g., x n Perform bitwise addition.

[0087] The cross-multiplication and bitwise addition described above can be applied to the fusion of data feature vectors with the same vector dimension (or number of elements). For data feature vectors with different vector dimensions, fusion can be performed after aligning the feature values ​​or padding with zeros.

[0088] Here, the obtained cross-feature vectors can be further fused. By performing the above cross-combination operation multiple times, the resulting cross-feature vectors continuously enrich the dimensions of the gene value feature vectors, enhancing their expressive power. Thus, by comprehensively utilizing methods such as feature cross-multiplication and positional addition, multiple rounds of processing can be applied to feature vectors of different types of data to continuously uncover potential relationships between features, enrich the dimensions of gene value feature vectors, and enhance their expressive power. Through complex and multi-round cross-combination operations, complex potential relationships in the data are continuously uncovered, enabling gene value feature vectors to more comprehensively and accurately reflect the intrinsic characteristics of the data. This provides more valuable input for subsequent feature calculation models (such as gene value calculation models), thereby improving the accuracy and reliability of gene value calculation to meet the needs of complex and ever-changing subject data feature analysis.

[0089] In step S420, the information contribution of each data feature vector and / or each cross feature vector in the target medical research data can be determined.

[0090] Here, information contribution refers to the importance of a feature vector (e.g., a data feature vector or a cross feature vector) in expressing the information of the target medical subject data.

[0091] As an example, information contribution can be represented by information gain, which can be used to quantify the information contribution of a feature. Specifically, suppose a dataset D with feature vectors (e.g., data feature vectors or cross feature vectors) is divided by feature A into V distinct values ​​a1, a2, ..., a... V D v In D, A takes the value a. v If the sample subset is a subset of the sample, then the information gain IG(D,A) can be expressed by the following equation (4):

[0092]

[0093] The information entropy H(D) can be represented by the following equation (5):

[0094]

[0095] Among them, C k Let K be the set of samples belonging to class k in dataset D, where K is the total number of sample classes. By calculating the information gain of each feature, we can evaluate its contribution to the classification of the dataset. The greater the information gain, the more important the feature is for classification (in the context of gene value calculation, this can be understood as distinguishing data from different topics).

[0096] Here, the information contribution can be used to determine the weight of the corresponding feature vector, and then fused with the corresponding feature vector in a weighted manner. Based on the weighted feature vector, the target data features can be determined, for example, and then input into the feature calculation model after fusion (which will be described in detail later).

[0097] By calculating the information contribution as described above, the contribution of features to the classification of research data (e.g., distinguishing between original and plagiarized content) can be quantified, thereby selecting the most discriminative features and those that best represent the uniqueness of the current research data, and optimizing the quality of gene value feature vectors. The input for calculating information contribution can include initial feature vectors (including feature vectors from structured, semi-structured, and unstructured data; that is, determining the information contribution of each feature vector in the target medical research data).

[0098] As an example, in step S420, the information contribution of each cross feature vector can also be determined. For example, the input for calculating the information contribution can also include the cross feature vector after multimodal fusion.

[0099] In step S430, the target data features can be determined based on the data feature vector, cross feature vector, and information contribution.

[0100] In this step, in one example, the target data features can be determined directly based on the data feature vector, cross feature vector, and information contribution.

[0101] In another example, data feature vectors can be filtered to obtain target feature vectors. Based on the target feature vectors, cross feature vectors, and the information contribution of each feature vector, the target data features are determined. For example, such as... Figure 5 As shown, step S430 may include:

[0102] In step S510, the correlation information between each data feature vector and the medical topic data type can be determined according to the preset medical topic data type.

[0103] Relevance information characterizes the relevance of data feature vectors in distinguishing different types of medical research topics. Here, the type of medical research topic can be determined based on the target medical research data. For example, it can be determined based on the research content, which may include the medical discipline to which the research belongs (e.g., basic medicine, clinical medicine, dentistry, etc.) or the research object (e.g., epidemiology, preventive medicine, internal medicine, surgery, etc.). Alternatively, it can be set based on whether the research data is original, i.e., original research, plagiarized research, etc. By analyzing the data feature vectors, the type of the current target medical research data can be determined, and different data feature vectors have different degrees of relevance in distinguishing types.

[0104] As an example, the mutual information between features can be used as correlation information. Specifically, let random variable X represent a feature, and Y represent a medical topic data type (e.g., category label). The mutual information MI(X;Y) can be represented by the following equation (6):

[0105]

[0106] Where, χ and Let X and Y be the value spaces of X and Y, respectively. p(x,y) is the joint probability of X=x and Y=y, and p(x) and p(y) are the marginal probabilities of X=x and Y=y, respectively. Based on the value calculated by the above formula (6), the correlation between each data feature vector and each preset medical topic data type is calculated.

[0107] Here, by calculating relevance information, the correlation between each data feature vector and each preset medical research topic data type can be determined. This is used to evaluate the correlation between features and data types and to select key features. Relevance information is a crucial component of the adaptive feature selection mechanism. It quantifies the non-linear dependency between features and research topic data category labels (e.g., originality or plagiarism), thereby selecting the features most discriminative for classification. By quantifying the dependency between features and category labels, relevance information provides a feature selection method that is closer to the semantic essence for gene value calculation. In complex scenarios involving multimodal data fusion, its synergistic effect with information contribution can significantly improve the quality of feature vectors, thereby enhancing the ability of gene values ​​to discriminate the originality of research topic data and providing more reliable technical support for the confirmation of scientific research data ownership.

[0108] In step S520, the target feature vector can be selected from the data feature vector based on the correlation information.

[0109] Here, the target feature vector can be any data feature vector that has a higher correlation with the data type of the medical topic than a preset level. Specifically, there can be multiple data feature vectors, and features with high correlation with the data type of the medical topic can be selected from these data feature vectors to construct the gene value feature vector.

[0110] As an example, the preset level can be determined according to actual needs. For example, a threshold can be preset for the calculated value of the above formula (6). When the calculated value of the correlation information of the data feature vector is greater than the threshold, it is considered that it has a high correlation with the data type of medical topic and has strong feature representativeness. Thus, feature vectors with low correlation with the data type of medical topic can be eliminated. For example, in response to the fact that the correlation between the data feature vector and all data types of medical topic is less than the preset threshold, or the number of data types of medical topic with a correlation less than the preset threshold is greater than the preset number, the data feature vector can be eliminated.

[0111] In step S530, the target data features can be determined based on the target feature vector, the cross feature vector, and the information contribution corresponding to the target feature vector and / or the cross feature vector.

[0112] Returning to step S430, in one example, the information contribution rate can be used to weight the corresponding feature vectors to obtain a weighted feature vector. This weighted feature vector is then used as the target data feature. Here, in examples where the data feature vectors are not filtered, the information contribution rate can be used to weight the corresponding data feature vectors and / or cross feature vectors; in examples where the data feature vectors are filtered, the information contribution rate can be used to weight the corresponding target feature vectors and / or cross feature vectors. Through this method, target data features can be extracted from the target medical research data, which can then be used for data feature comparison.

[0113] In this example, the similarity between target data from different topics can be determined by calculating the feature distance (e.g., Euclidean distance, Manhattan distance, norm distance, etc.) between the target data features. In this case, the preset level for similarity can be a preset value for the feature distance, which can be set according to actual needs.

[0114] In another example, to further improve the accuracy of data feature comparison, a feature calculation model can be pre-trained to calculate the quantified values ​​of data features. The similarity between the two data can be determined by subtracting the quantified values ​​of data features from different topics.

[0115] For example, the target data feature can be a gene value. In this example, step S530 may include: inputting the target feature vector, the cross feature vector, and the information contribution corresponding to the target feature vector and / or the cross feature vector into a pre-trained feature calculation model, and outputting the target data feature from the feature calculation model.

[0116] Here, the feature calculation model may include a residual network, an attention module, and a prediction module (e.g., ...). Figure 7 The architecture shown is as follows: Figure 6 As shown, the feature calculation model can be trained in the following ways:

[0117] In step S610, a training sample set of medical research data can be obtained, wherein the training sample set includes multiple training samples and sample labels for each training sample.

[0118] Here, each training sample may include multiple target feature vectors, sample cross feature vectors, and sample information contribution values ​​corresponding to each target feature vector and / or sample cross feature vector. The determination process for the target feature vectors, sample cross feature vectors, and sample information contribution values ​​is the same as that described above, and therefore will not be repeated here.

[0119] Sample labels can represent the gene values ​​of training samples. As an example, gene values ​​can be a quantitative representation of global features of medical research data, used to uniquely identify the data content, and their calculation can rely on the fusion results of multi-dimensional features of the data.

[0120] As an example, before model training, gene values ​​can be pre-generated for training samples using benchmark methods, serving as labels for supervised learning. For instance, a benchmark method could be to use a hash function (e.g., the SHA-256 algorithm) to calculate a fixed-length hash value on the fused composite feature vector. Due to the characteristics of hash functions, subtle changes in the data can lead to significantly different hash values; this hash value then serves as the gene value for the medical research data, uniquely identifying the data content. In step S620, a residual network can be used to extract the semantic information of the target feature vector for each training sample.

[0121] Here, the target feature vector of the training samples can be input into the residual network of the feature calculation model, and the semantic information can be extracted using the residual network.

[0122] As an example, a residual network can be a deep residual neural network, which can include multiple convolutional layers. The output of the residual network can be the sum of the input sample target feature vector and the output vector of the last convolutional layer.

[0123] Taking two convolutional layers as an example, we can assume the input feature vector is... It passes through two convolutional layers in sequence, where the output of the first convolutional layer can be... The output of the second convolutional layer can be Where W1 and W2 are the convolution kernel weight matrices, b1 and b2 are the bias vectors, and σ is the activation function (e.g., ReLU), then the output of the residual network... It can be represented as:

[0124] This residual connection method can solve the problems of gradient vanishing and gradient exploding, enabling the network to learn the residual mapping between input and output. As the number of network layers increases, it can better mine the high-order semantic information in the feature vector of the subject data.

[0125] In step S630, the target feature vector of each training sample can be input into the attention module to obtain the attention fusion feature vector corresponding to each training sample.

[0126] In this step, an attention mechanism can be introduced for fusion, extracting the weights of the target feature vectors of each sample to highlight key information. The target feature vectors of the training samples can be input into the attention module, and the attention module outputs the processed vector.

[0127] As an example, we can assume the input feature vector is... The attention mechanism calculates the weight α for each feature dimension. i To highlight key information, a query vector can be obtained through linear transformation. key vector Sum value vector They are respectively: Among them, W q W k W v It is a learnable weight matrix.

[0128] Based on the query vector and key vector, the attention score e can be calculated. ij : The attention weight α can be obtained by performing Softmax normalization on the attention score. ij :

[0129] Based on the attention weights and value vectors, the output vector after processing by the attention mechanism is obtained. It can be represented as:

[0130] In step S640, the prediction module can be used to predict the gene value of each training sample based on the semantic information of the target feature vector of each sample in each training sample, the attention fusion feature vector, the sample cross feature vector and the information contribution, so as to obtain the prediction result.

[0131] The semantic information, attention fusion feature vector, sample cross-feature vector, and information contribution of each training sample's target feature vector can be input into the prediction module to obtain the gene value of the training sample output by the prediction module, which serves as the prediction result. As an example, the prediction module can be a graph neural network (GNN) to better adapt to the need to handle the complex correlations of multimodal features in medical research data (such as the cross-influence of experimental parameters and textual descriptions).

[0132] In step S650, the feature calculation model can be trained based on the prediction results and sample labels to obtain a trained feature calculation model.

[0133] In this step, the predicted results can be compared with the sample labels to train the model. For example, the model parameters (such as the residual network, attention module, and weights of the prediction module) can be optimized through backpropagation to minimize the error (such as mean squared error) between the predicted values ​​and the baseline labels.

[0134] As an example, according to an embodiment of this disclosure, after the model is trained and embedded into the project evidence storage system for online application, the model can be continuously updated based on online data to achieve continuous learning of the model.

[0135] For online learning and updating of the model, in one example, stochastic gradient descent (SGD) can be used to update the model parameters. Specifically, let the model parameters be θ and the loss function be L(θ). For a newly added batch of samples S, the SGD algorithm can update the parameters using the following equation (7):

[0136]

[0137] Where η is the learning rate. This indicates calculating the gradient with respect to the parameter θ. For example, when new research data is added, the gradient is calculated using this data, and the model parameters are updated according to the above formula, enabling the model to adapt to the new data and maintain its adaptability and computational accuracy to the latest research data.

[0138] In another example, the Adagrad algorithm can also be used to update the model's parameters. Specifically, the Adagrad algorithm adjusts the learning rate based on the gradient history of the parameters. For example, we can set g... τ,i The parameter θ is the parameter at the t-th iteration. i The gradient of G t,ii It is a diagonal matrix whose diagonal elements are the parameters θ up to the t-th iteration. i The sum of squares of the gradient, i.e. In this case, the Adagrad algorithm can update the model's parameters θ using the following equation (8). i :

[0139]

[0140] Where ∈ is a very small positive number (e.g., 10). -8 ), used to prevent the denominator from being zero.

[0141] After the model is trained, the feature vector set of the research data for which gene values ​​are to be calculated can be input into the trained model to output gene values. Here, the gene values ​​output by the model can be directly used for data uniqueness identification, without relying on benchmark methods to determine gene values. Through the above logic, the model can fit pre-generated gene value labels through supervised learning.

[0142] Return to reference Figure 2 In step S240, in response to the comparison result indicating that the similarity between the target data features and the data features of all stored medical research data is less than a preset level, the research evidence system determines the evidence information corresponding to the target medical research data and stores the evidence information in the blockchain.

[0143] Specifically, after a user submits target medical research topic data, the target medical research topic data is used in the research topic storage system to calculate its target data features (e.g., gene values). The system then determines whether the similarity between the target data features and existing data features is higher than a preset level (e.g., gene similarity > 80%), or whether the target data features already exist in the feature library (e.g., gene similarity = 100%).

[0144] If the similarity to all existing data features is less than or equal to a preset level (e.g., gene similarity less than or equal to 80%), the target medical research data can be considered free from infringement risks, and the data can be stored and published.

[0145] Here, the project evidence storage system can determine the evidence storage information based on the target medical project data. The evidence storage information can include information on the characteristics of the target data and the upload time of the target medical project data.

[0146] As an example, such as Figure 3 As shown in the example, the project evidence storage system can calculate the hash value of the target data features (such as gene value data) and use it as information about the target data features. Furthermore, the system can bind the target medical project data to an authoritative time source to generate a trusted timestamp. For example, the National Time Service Center can be responsible for time service and adherence, and a trusted third-party timestamp service provider can digitally sign the electronic data summary and authoritative time record to generate the trusted timestamp. The trusted timestamp can be used to determine the precise time when the target medical project data, as an electronic document, was generated, preventing tampering and providing credible time proof and proof of the authenticity and integrity of the content, making it authoritative and trustworthy as evidence. As an example, the evidence storage information can also include a digital certificate for the target medical project data. For instance, based on the aforementioned trusted timestamp, an interface such as the China Financial Certification Authority (CFCA) can be called to obtain its issued digital certificate (or electronic ID card) to ensure that the content of the target medical project data has not been tampered with and that the signature has not been altered, thereby ensuring the integrity and authenticity of the ownership information.

[0147] After determining the aforementioned evidence storage information (or evidence storage message information), the project evidence storage system, if... Figure 3As shown, evidence information can be stored in a blockchain system, and an electronic data evidence certificate can be generated. Furthermore, as an example, the project evidence storage system can also simultaneously synchronize the evidence information to the internet court's consortium blockchain, thereby providing strong and effective legal protection for original works or achievements. After successfully uploading to the consortium blockchain, the transmission information synchronized to the consortium blockchain can be returned to the blockchain platform, such as returning a hash value, so that the transmission information can also be stored on the blockchain, facilitating the recording of the entire process for traceability and subsequent querying on the blockchain traceability information page for rights confirmation.

[0148] As an example, the project evidence storage system can also store the evidence information and the target medical project data in a local database.

[0149] In the embodiments of this disclosure, in response to the comparison result indicating that the similarity between the data features of the target medical research data and the data features of any stored medical research data is greater than a preset level, the research evidence storage system can send an infringement warning message to the data platform. The infringement warning message is used to remind the uploader of any stored data that there is a risk that any stored data has been infringed.

[0150] like Figure 3 As shown, if the similarity exceeds a preset level, such as a gene similarity exceeding 80%, the target medical research data poses a risk of infringement and is considered suspected infringing data. In such cases, data submission cannot be completed, and the user is notified of submission failure and the reason for the failure. In this situation, the data platform can store information about the failed submission of the medical research data, for example, in a local database.

[0151] By using the methods described above, we can prevent medical research data that poses a risk of infringement from being published on data platforms, thus protecting the rights of those who upload published medical research data.

[0152] The following describes an example of the user registration process in a medical research data storage system. Before a user uploads target medical research data, in this embodiment of the disclosure, the data platform can respond to a user registration request sent by a first terminal by obtaining the user information of the user to be registered, wherein the user information includes iris information and fingerprint information; the data platform can generate iris feature private key and iris feature public key corresponding to the iris information, and fingerprint feature private key and fingerprint feature public key corresponding to the fingerprint information, and store the iris feature private key, iris feature public key, fingerprint feature private key and fingerprint feature public key in a local database and generate a storage record.

[0153] In addition, the data platform can store iris feature public keys and fingerprint feature public keys on the blockchain, receive transaction information returned from the blockchain, and store the transaction information and corresponding storage records in a local database.

[0154] Specifically, with Figure 3 For example, researchers need to submit their research data to a data platform within a specified timeframe. Using this platform requires registration of personal information, such as basic information, and the extraction of iris and fingerprint features using iris and fingerprint recognition technologies, respectively. The data platform generates private keys from the iris and fingerprint features and uses an elliptic curve algorithm to generate corresponding public keys for the iris and fingerprint features. The basic information, iris and fingerprint features, along with their respective private and public keys, are stored together in a local database. After successful database storage, the data platform can store the iris and fingerprint public keys in a blockchain system. The blockchain system can then return transaction information (such as a blockchain transaction ID), and the data platform can update the corresponding records in its local database with the received transaction information.

[0155] In addition, as an example, besides uploading and publishing medical research data as described above, the research data storage method based on the embodiments of this disclosure can also query (or retrieve) and reference published medical research data.

[0156] For example, the data storage method may further include: In response to receiving a query request from a second terminal for the target medical topic data, the data platform sends the queried target medical topic data to the second terminal; In response to receiving a citation request from the second terminal for the queried target medical topic data, the data platform queries the blockchain for the first public key of the target medical topic data stored on the blockchain, and compares the first public key with the second public key of the target medical topic data stored on the data platform; In response to the first public key and the second public key matching, the data platform verifies the signature information of the target medical topic data; In response to the first public key and the second public key not matching, the data platform provides the second terminal with a warning message indicating that the target medical topic data may have been tampered with. Here, the first public key and the second public key can indicate the identity information of the uploader of the target medical topic data.

[0157] Specifically, the second terminal can be a user terminal, which may be the same as or different from the first terminal. Users of the second terminal can log in to the data platform as users. After logging in to the data platform, users can search for published topics and view the data of those topics. For example, they can search for and view the target medical topic data published through the above method.

[0158] Users can reference searched data on target medical topics. When referencing this data, the data platform can check whether the public key stored in the database matches the public key stored in the blockchain. If they don't match, the data is considered untrustworthy, and the user is alerted that the data may have been tampered with. This message is also sent to the data provider and the administrator of the data storage system. If they match, the signature information of the data can be verified.

[0159] By querying and referencing data in the above manner, the authenticity and reliability of the data can be ensured, and the tampering of medical research data can be prevented.

[0160] Furthermore, as an example, when a user retrieves the data, the data platform can also perform the aforementioned step of querying whether the characteristic public key of the medical topic data stored in the database is consistent with the characteristic public key stored in the blockchain. That is, both the retrieved data and the referenced data are verified to further ensure the authenticity and reliability of the data.

[0161] Specifically, in response to receiving a query request for target medical topic data from a second terminal, before sending the queried target medical topic data to the second terminal, the data platform queries the blockchain for a first feature public key and compares the first feature public key with a second feature public key. If the first feature public key and the second feature public key match, the data platform verifies the signature information of the target medical topic data. If the first feature public key and the second feature public key do not match, the data platform sends a warning message to the second terminal indicating that the target medical topic data may have been tampered with.

[0162] According to embodiments of this disclosure, the step of verifying the signature information of target medical research data may include: a data platform querying the blockchain for first signature information of the target medical research data stored on the blockchain, and comparing the first signature information with second signature information of the target medical research data stored on the data platform; in response to the first signature information and the second signature information being consistent, the data platform verifying the second signature information using a second feature public key; and in response to successful verification of the second signature information, the data platform sending the target medical research data to a second terminal. Here, the first signature information and the second signature information can be generated based on a feature private key corresponding to the feature public key of the target medical research data.

[0163] Specifically, such as Figure 8As shown, the signature information stored in the database can be compared with the signature information stored in the blockchain through the transaction information of the blockchain. If they do not match, the signature information of the medical project data is untrustworthy, and the user is notified that the data signature is untrustworthy and may have been tampered with. This message is also sent to the medical project data provider and administrator. If they match, the signature information is verified using an elliptic curve algorithm with either an iris feature public key or a fingerprint feature public key. If the verification fails, it indicates that the data has been tampered with and is untrustworthy. If they match, it means that the medical project data is normal and has not been tampered with, and the plaintext data result is provided to the client for use.

[0164] By using the above methods, in addition to local and blockchain comparison and verification of the feature public key, the signature information can be further compared and verified locally and on the blockchain. This allows for verification from two dimensions: the feature information of the medical research data itself and the signature information of the data publisher, thus achieving dual verification of the data and further improving data reliability.

[0165] Once validation is successful, users can cite data related to the target medical topic. Furthermore, as an example, such as... Figure 8 As shown, after the target medical research data is successfully cited, the data platform can store citation information, including the citer information, the medical research data, and the cumulative number of citations, in the local database. After successful storage, the citation information can also be stored in the blockchain system. The blockchain system returns a transaction ID, and the transaction ID information is updated in the corresponding record in the saved local database.

[0166] The above description describes the ability to compare the currently uploaded target medical research data with stored data in the database for infringement detection. As an example, the evidence storage method according to an embodiment of this disclosure may further include: the research data storage system queries target data in a preset target resource at preset time intervals, and compares the stored medical research data with the target data for data features; in response to the comparison result indicating that the similarity of the data features between the stored medical research data and the target data is greater than a preset level, it can be considered that an infringement risk has been discovered, a warning message indicating the risk of infringement is sent to the uploader of the stored medical research data, and the updated data is stored as evidence; in response to the comparison result indicating that the similarity of the data features between the stored medical research data and the target data is less than or equal to a preset level, it can continue to wait for the next time interval to arrive.

[0167] Specifically, such as Figure 9As shown, the research project data storage system can be configured with scheduled tasks that automatically detect preset target resources and perform data capture and analysis on medical research projects. Here, target resources can be, for example, publicly available resources on a specified webpage or public resource library, and target data can be, for example, resources updated on that webpage or public resource library within the time period from the last execution of the scheduled task to the current time. Furthermore, the timing can be set according to actual needs; for example, it can be to execute the scheduled task at preset time intervals (e.g., weekly, monthly), and / or on pre-specified dates.

[0168] Here, as Figure 9 As shown, data features can be, for example, gene values. The process of comparing the stored medical research data with the target data using data features is the same as the process described above of comparing the target medical research data with the medical research data stored in the database using data features, and will not be repeated here.

[0169] Furthermore, according to embodiments of this disclosure, the subject matter evidence storage system can also, in response to receiving an infringement detection instruction provided by a user, compare the data of the target page information (e.g., suspected infringing web page information) in the infringement detection instruction with the subject matter data uploaded by the user.

[0170] Furthermore, according to embodiments of this disclosure, the project evidence storage system can also, in response to receiving suspected infringement materials provided by a user, compare the data in the suspected infringement materials with the project data uploaded by the user.

[0171] Whether comparing target medical research data with existing medical research data in the database, or comparing existing medical research data with target data, user-provided pages, or suspected infringing materials, if a user submits suspected infringing / plagiarized data, or if the research data storage system itself searches and analyzes suspected infringing / plagiarized data (e.g., calculating a similarity of data features >80%), the system will confirm this event as a suspected infringement event. The system can send a notification message to the contact information pre-maintained by the user who uploaded the research data at risk of infringement.

[0172] Furthermore, whether comparing the target medical research topic data with existing medical research topic data in the database, or comparing existing medical research topic data with the target data, user-provided pages, or suspected infringing materials, the research topic evidence storage system, in response to the discovery of infringement risks, not only issues warnings to users but also stores evidence of infringement, such as... Figure 9As shown, it allows for the storage of solidified evidence such as webpage screenshots, webpage source code downloads (screenshots), and webpage service information collection. Furthermore, it enables the calculation of hash values ​​for this key information and its storage on a blockchain system, which can then be synchronized with authoritative institutions (such as one or more judicial bodies), including internet courts. Upon receiving an infringement warning, users can file a rights protection application with the judicial body, which can then retrieve the solidified evidence uploaded to the blockchain's evidence storage system.

[0173] According to the data evidence preservation method of the embodiments of this disclosure, a data reading and storage encryption scheme for a medical research data platform and a blockchain system can be designed. Combined with blockchain technology, this ensures data security and traceability. This method can achieve second-level evidence preservation, enabling online appeals for infringement incidents. The system automatically acquires original works with confirmed copyright on web pages and automatically scans suspected plagiarized works, providing an immutable copyright evidence preservation system and blockchain-based traceability records for each step. Furthermore, this method can synchronize the evidence preservation information to authoritative institutions (such as one or more judicial institutions), providing strong and effective legal protection for original works and issuing a certificate of ownership and a unique identifier for each work.

[0174] Furthermore, the data evidence preservation method according to the embodiments of this disclosure can strengthen the information verification during the admission of researchers, ensure the accuracy of data cited by users of medical research data, and provide researchers and data users with traceability and query of the entire chain of medical research data publication, citation, rights confirmation and protection and evaluation, avoiding or at least alleviating the problems of difficult and easily tampered data evidence preservation, and to a certain extent ensuring the rigor and reliability of medical research data.

[0175] Furthermore, the data storage method according to the embodiments of this disclosure can obtain basic project information, project data information, citation count, citation information, rights confirmation, rights protection, and other related information of the project data by querying the transaction ID information of medical project data on the blockchain. It can also display the entire chain of traceability information related to the project, achieving reliable full-chain traceability management of project data. Here, traceability refers to the fact that the blockchain stores all historical data after the genesis block through a block data structure, and any piece of data on the blockchain can be traced back to its origin through the chain structure.

[0176] Furthermore, the data storage method according to the embodiments of this disclosure can combine multiple technical solutions, such as blockchain technology, electronic signatures, trusted timestamps, data storage schemes, and data retrieval schemes. This ensures the credibility of personal information and background check information, preventing data tampering, while also reducing the cost of storage and traceability, and achieving end-to-end traceability management of medical research data storage and publication. Here, immutability (Tamper Proof) means that once information on the blockchain is agreed upon and added to the blockchain, it is jointly recorded by all nodes, and cryptographically ensures the interconnectivity of each subsequent record, making tampering extremely difficult and costly.

[0177] According to a second aspect of this disclosure, a medical research data storage system is provided, which includes a data platform, a research data storage system, and a local database.

[0178] Upon receiving the target medical research data uploaded by the first terminal, the data platform sends the target medical research data to the research data storage system.

[0179] The project evidence storage system encodes the target medical project data according to the data structure type, and obtains the data feature vector corresponding to the data of each data structure type in the target medical project data.

[0180] The project evidence storage system obtains target data features by fusing data feature vectors and compares the target data features with the data features of medical project data already stored in the local database.

[0181] In response to the comparison results indicating that the similarity between the target data features and the data features of all stored medical research data is less than a preset level, the research project evidence storage system determines the evidence information corresponding to the target medical research data and stores the evidence information on the blockchain. The evidence information includes the target data feature information and the upload time of the target medical research data.

[0182] Regarding the medical research data storage system in the above embodiments, the structure of each part and the specific method of execution have been described in detail in the embodiments of the medical research data storage method, and will not be elaborated here.

[0183] The following is an example of the system architecture for implementing this medical data storage system. The system may include a blockchain layer, a distributed storage layer, an encryption and privacy protection layer, and a user interaction layer.

[0184] As an example, the blockchain layer can store metadata of the research data. Metadata may include data hashes, descriptive information, Merkle Tree root nodes, etc. Here, only metadata is stored on the chain, which can reduce the storage pressure on the blockchain.

[0185] Furthermore, blockchain can implement smart contracts for permission management and access control. Here, the hash value of data can be used to verify data integrity, and the data description information can include data type, source, etc. The Merkle Tree root node can store the research data in blocks, with each block having its hash value calculated, organized into a Merkle Tree. On the blockchain, only the root node can be stored, and users can efficiently verify data integrity through off-chain data blocks and paths. Here, the blockchain can be implemented, for example, through Ethereum or Hyperledger Fabric, to record metadata and implement smart contracts. Blockchain and Merkle Tree ensure data immutability and integrity.

[0186] As an example, the local database can be a distributed storage system, such as the InterPlanetary File System (IPFS), to store encrypted research data. This database can also provide data links, associating it with off-chain data through on-chain metadata records. For instance, research data can be encrypted and stored in a local database (e.g., IPFS), generating a unique Content Identifier (CID). This CID, along with the hash value of the encryption key, can be recorded on the blockchain.

[0187] As an example, in the encryption and privacy protection layer, encryption methods for research data can include homomorphic encryption, such as Paillier and CKKS algorithms, to ensure data privacy and secure transmission. Using homomorphic encryption allows users to perform calculations on encrypted research data, meeting research needs. For example, users can submit computation tasks off-chain and receive the results without decrypting the data.

[0188] As an example, in the process of transmitting medical research data between the data storage system and the blockchain, encrypted data transmission can be used. After the user is verified, the off-chain service will securely transmit the encrypted data to the user. For example, protocols such as Transport Layer Security (TLS) can be used for data transmission to ensure the security of off-chain data.

[0189] As an example, a medical data storage system can employ smart contracts to record users' permission status. This permission status can include information about the data owner (uploader) and the data accessor's permission status (e.g., whether they have permission to access certain data). By using smart contracts, dynamic permission updates are also supported, allowing the data owner to revoke or grant permissions at any time. After a user initiates an access request to the system, the smart contract verifies their permissions. Upon successful verification, an off-chain service retrieves the encrypted data based on the CID and transmits it to the user.

[0190] As an example, a medical research data storage system may also include a user interface that allows users to submit data, request access permissions, and verify data integrity, and may also include off-chain services such as data transfer and decryption services.

[0191] The following describes an example of the data upload process for this medical research data storage system. The data owner can divide the research data into blocks and calculate a hash value for each block to generate a Merkle Tree. Then, homomorphic encryption can be used to encrypt the data, which can then be uploaded to IPFS, and a CID can be obtained. Metadata, including but not limited to data description information, data hash values, the Merkle Tree root node, and the IPFS CID, can be recorded on the blockchain.

[0192] The following describes an example of the data access process for this medical research data storage system. Users can initiate an access request to the smart contract, providing their identity information. The smart contract verifies user permissions: if the user has permission, access is granted; if not, access is denied. Off-chain services can retrieve encrypted data from IPFS based on the CID and securely transmit it to the user. Users verify data integrity through the Merkle Tree root node recorded on-chain. Users can perform computations on the data off-chain or execute computational tasks using homomorphic encryption.

[0193] The following describes an example of the data verification process for this medical data storage system. After a user obtains a data block, data integrity is verified using a Merkle Tree: the root node is calculated using the hash value and path of the data block. The calculated result is compared with the root node stored on the chain. If the verification passes, data integrity is guaranteed.

[0194] Based on the above system architecture, blockchain, distributed storage, and encryption technologies can be combined to ensure data security and privacy while meeting the needs of scientific research data sharing and computation.

[0195] According to a third aspect of this disclosure, a computer-readable storage medium is provided that, when instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform a blockchain-based method for storing medical research data according to this disclosure.

[0196] Here, the electronic device may include a processor and a memory for storing processor-executable instructions. When executed by the processor, the processor-executable instructions cause the processor to perform the training method or adversarial method for adversarial clustering as described in the exemplary embodiments above.

[0197] As an example, an electronic device does not necessarily have to be a single device; it can also be a collection of any means or circuits capable of executing the aforementioned instructions (or instruction sets) individually or in combination. An electronic device can also be part of an integrated control system or system manager, or can be configured to interconnect with a server, either locally or remotely (e.g., via wireless transmission), through an interface.

[0198] In electronic devices, processors may include central processing units (CPUs), graphics processing units (GPUs), programmable logic devices, dedicated processor systems, microcontrollers, or microprocessors. By way of example and not limitation, processors may also include analog processors, digital processors, microprocessors, multi-core processors, processor arrays, network processors, etc.

[0199] The processor can execute instructions or code stored in memory, which can also store data. Instructions and data can also be sent and received over a network via a network interface device, which can employ any known transport protocol.

[0200] Memory can be integrated with the processor; for example, RAM or flash memory can be housed within an integrated circuit microprocessor. Alternatively, memory can comprise a separate device, such as an external disk drive, storage array, or other storage device that can be used by any database system. Memory and processor can be operatively coupled, or can communicate with each other, for example, via I / O ports, network connections, etc., enabling the processor to read files stored in the memory.

[0201] In addition, electronic devices may include video displays (such as liquid crystal displays) and user interaction interfaces (such as keyboards, mice, touch input devices, etc.). All components of the electronic device can be interconnected via buses and / or networks.

[0202] Computer-readable storage media may be, for example, a memory that includes instructions. Optionally, computer-readable storage media may be: read-only memory (ROM), random access memory (RAM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disc storage, hard disk drive (HDD), solid-state drive (SSD), card storage (such as multimedia cards, secure digital (SD) cards, or ultra-fast digital (XD) cards), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, and any other device configured to store a computer program and any associated data, data files, and data structures in a non-transitory manner and to provide the computer program and any associated data, data files, and data structures to a processor or computer so that the processor or computer can execute the computer program. The computer program in the aforementioned computer-readable storage medium can run in an environment deployed in computer devices such as clients, hosts, agent devices, servers, etc. Furthermore, in one example, the computer program and any associated data, data files, and data structures are distributed across a networked computer system, such that the computer program and any associated data, data files, and data structures are stored, accessed, and executed in a distributed manner through one or more processors or computers.

[0203] According to a fourth aspect of this disclosure, a computer program product is provided, including computer-executable instructions that, when executed by at least one processor, implement the blockchain-based medical subject data storage method according to this disclosure.

[0204] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0205] Furthermore, it should be noted that although several examples of each step have been described above with reference to the specific accompanying drawings, it should be understood that the embodiments of this disclosure are not limited to the combinations given in the examples. The steps appearing in different drawings can be combined, and the execution order of each step can be changed, which will not be exhaustive here.

[0206] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for storing medical research data based on blockchain, characterized in that, The method is applied to a medical research project data storage system, which includes a data platform, a research project storage system, and a local database. The method includes: In response to receiving the target medical research data uploaded by the first terminal, the data platform sends the target medical research data to the research evidence storage system. The project evidence storage system encodes the target medical project data according to the data structure type, and obtains data feature vectors corresponding to the data of each data structure type in the target medical project data; The project evidence storage system obtains target data features by fusing the data feature vectors, and compares the target data features with the data features of medical project data already stored in the local database. In response to the comparison result indicating that the similarity between the target data features and the data features of all stored medical research data is less than or equal to a preset level, the research project evidence storage system determines the evidence storage information corresponding to the target medical research data and stores the evidence storage information on the blockchain. The evidence storage information includes information about the target data features and the upload time of the target medical research data. The project evidence storage system obtains the target data features in the following ways: At least one fusion is performed between data feature vectors corresponding to different data structure types to obtain cross feature vectors, wherein the preset data structure types include structured data, semi-structured data, and unstructured data; Determine the information contribution of each data feature vector and / or each cross feature vector in the target medical research data, wherein the information contribution represents the importance of the data feature vector to expressing the information of the target medical research data; Based on the data feature vector, the cross feature vector, and the information contribution degree, the target data features are determined. The step of determining the target data features based on the data feature vector, the cross feature vector, and the information contribution degree includes: Based on the preset medical topic data types, determine the correlation information between each data feature vector and the medical topic data types; Based on the correlation information, a target feature vector is selected from the data feature vector, wherein the target feature vector is a data feature vector in the data feature vector whose correlation with the data type of the medical topic is higher than a preset level; The target data features are determined based on the target feature vector, the cross feature vector, and the information contribution corresponding to the target feature vector and / or the cross feature vector.

2. The data storage method according to claim 1, characterized in that, The target data feature is a gene value, wherein determining the target data feature based on the target feature vector, the cross feature vector, and the information contribution of the target feature vector includes: The target feature vector, the cross feature vector, and the information contribution of the target feature vector are input into a pre-trained feature calculation model, and the target data features are output from the feature calculation model. The feature calculation model includes a residual network, an attention module, and a prediction module, and is trained in the following manner: A training sample set for medical research data is obtained, wherein the training sample set includes multiple training samples and sample labels for each training sample, wherein each training sample includes multiple sample target feature vectors, sample cross feature vectors, and information contribution of each sample target feature vector, and the sample label represents the gene value of the training sample; Using the residual network, the semantic information of the target feature vector of each training sample is extracted; The target feature vector of each training sample is input into the attention module to obtain the attention fusion feature vector corresponding to each training sample. Using the prediction module, the gene value of each training sample is predicted based on the semantic information, attention fusion feature vector, sample cross feature vector, and information contribution of the target feature vector of each sample in each training sample, and the prediction result is obtained. Based on the prediction results and the sample labels, the feature calculation model is trained to obtain a trained feature calculation model.

3. The data storage method according to claim 1, characterized in that, The data preservation method also includes: In response to receiving a query request from a second terminal for the target medical topic data, the data platform sends the queried target medical topic data to the second terminal. In response to receiving a reference request from a second terminal for the queried target medical topic data, the data platform queries the blockchain for the first feature public key of the target medical topic data stored on the blockchain, and compares the first feature public key with the second feature public key of the target medical topic data stored on the data platform, wherein the first feature public key and the second feature public key indicate the identity information of the uploader of the target medical topic data; In response to the first feature public key being consistent with the second feature public key, the data platform verifies the signature information of the target medical research data; In response to the inconsistency between the first feature public key and the second feature public key, the data platform sends a warning message to the second terminal indicating that the target medical research data may have been tampered with.

4. The data storage method according to claim 3, characterized in that, The verification of the signature information of the target medical research data includes: The data platform queries the blockchain for the first signature information of the target medical research data stored on the blockchain, and compares the first signature information with the second signature information of the target medical research data stored on the data platform. In response to the consistency between the first signature information and the second signature information, the data platform uses the second feature public key to verify the second signature information; In response to successful verification of the second signature information, the data platform sends the target medical research data to the second terminal. The first signature information and the second signature information are generated based on the feature private key corresponding to the feature public key of the target medical research data.

5. The data storage method according to claim 3 or 4, characterized in that, The data preservation method also includes: In response to receiving a query request from a second terminal for the target medical topic data, the data platform queries the blockchain for the first feature public key before executing the step of sending the queried target medical topic data to the second terminal, and compares the first feature public key with the second feature public key. In response to the first feature public key being consistent with the second feature public key, the data platform verifies the signature information of the target medical research data; In response to the inconsistency between the first feature public key and the second feature public key, the data platform sends a warning message to the second terminal indicating that the target medical research data may have been tampered with.

6. A medical research data storage system, characterized in that, The medical research data storage system includes a data platform, a research data storage system, and a local database. In response to receiving the target medical research data uploaded by the first terminal, the data platform sends the target medical research data to the research evidence storage system. The project evidence storage system encodes the target medical project data according to the data structure type, and obtains data feature vectors corresponding to the data of each data structure type in the target medical project data; The project evidence storage system obtains target data features by fusing the data feature vectors, and compares the target data features with the data features of medical project data already stored in the local database. In response to the comparison result indicating that the similarity between the target data features and the data features of all stored medical research topics is less than a preset level, the research topic evidence storage system determines the evidence information corresponding to the target medical research topic data and stores the evidence information on the blockchain. The evidence information includes information about the target data features and the upload time of the target medical research topic data. The project evidence storage system obtains the target data features in the following ways: At least one fusion is performed between data feature vectors corresponding to different data structure types to obtain cross feature vectors, wherein the preset data structure types include structured data, semi-structured data, and unstructured data; Determine the information contribution of each data feature vector and / or each cross feature vector in the target medical research data, wherein the information contribution represents the importance of the data feature vector to expressing the information of the target medical research data; Based on the data feature vector, the cross feature vector, and the information contribution degree, the target data features are determined. The step of determining the target data features based on the data feature vector, the cross feature vector, and the information contribution degree includes: Based on the preset medical topic data types, determine the correlation information between each data feature vector and the medical topic data types; Based on the correlation information, a target feature vector is selected from the data feature vector, wherein the target feature vector is a data feature vector in the data feature vector whose correlation with the data type of the medical topic is higher than a preset level; The target data features are determined based on the target feature vector, the cross feature vector, and the information contribution corresponding to the target feature vector and / or the cross feature vector.

7. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the blockchain-based medical subject data storage method according to any one of claims 1 to 5.

8. A computer program product comprising computer-executable instructions, characterized in that, When the computer-executable instructions are executed by at least one processor, they implement the blockchain-based medical research data storage method according to any one of claims 1 to 5.

Images