Federal learning method and device based on special disease database

Abstract

The invention relates to the technical field of artificial intelligence, and relates to a federal learning method and device based on a special disease database, and the method comprises the steps: obtaining federal learning request information; determining a first node according to the federal learning request information, wherein the first node is a node for issuing a training task; training task information is obtained according to the first node, the training task information comprises a training task sent to at least one second node by the first node, and the second node is a node for executing the training task by the local end; constructing a special disease database at a local end based on the training task information to serve as training data to train a local model, and obtaining a trained local model; and sending the trained local model of each second node to the first node for aggregation to obtain a global model.According to the invention, the nodes on the block chain are selected as the first node and the second nodes, and the block chain and federated learning are combined, so that direct communication between participating nodes is realized, and the communication efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a federated learning method and device based on a disease-specific database. Background Art

[0002] Federated learning can effectively help multiple medical institutions' disease-specific databases to perform machine learning while ensuring privacy protection and data security, and enable collaborative computing among multiple medical structures without sharing the original disease-specific databases. However, for data sharing among multiple medical institutions, if malicious nodes are mixed into the participating nodes, privacy leaks may occur. Summary of the invention

[0003] The purpose of the present invention is to provide a federated learning method and device based on a disease-specific database to improve the above-mentioned problems.

[0004] In order to achieve the above objectives, the present application provides the following technical solutions: On the one hand, an embodiment of the present application provides a federated learning method based on a disease-specific database, the method comprising: Get federated learning request information; Determine a first node according to the federated learning request information, where the first node is a node that publishes a training task; Acquire training task information according to the first node, where the training task information includes a training task sent by the first node to at least one second node, where the second node is a node that executes the training task locally; Based on the training task information, a disease-specific database is constructed locally as training data to train the local model, thereby obtaining a trained local model; The trained local model of each second node is sent to the first node for aggregation to obtain a global model.

[0005] In a second aspect, an embodiment of the present application provides a federated learning device based on a disease-specific database, the device comprising: A first acquisition module is used to obtain federated learning request information; A first processing module, configured to determine a first node according to the federated learning request information, where the first node is a node that publishes a training task; A second acquisition module is used to acquire training task information according to the first node, wherein the training task information includes a training task sent by the first node to at least one second node, and the second node is a node that executes the training task locally; A second processing module is used to construct a disease-specific database on the local side based on the training task information as training data to train the local model, thereby obtaining a trained local model; The third processing module is used to send the trained local model of each second node to the first node for aggregation to obtain a global model.

[0006] In a third aspect, an embodiment of the present application provides a federated learning device based on a disease-specific database, the device comprising a memory and a processor. The memory is used to store a computer program; the processor is used to implement the steps of the above-mentioned federated learning method based on a disease-specific database when executing the computer program.

[0007] In a fourth aspect, an embodiment of the present application provides a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned federated learning method based on a disease-specific database are implemented.

[0008] The beneficial effects of the present invention are: The present invention first ensures the credibility of the node that publishes the training task by determining the first node as the node that publishes the training task. Then, the training task information is sent to multiple second nodes, so that they build a disease-specific database locally as training data to train the local model. This can effectively ensure data security and avoid data leakage. In addition, the present invention liberates the central server by combining blockchain and federated learning, realizes direct communication between participating nodes, and improves communication efficiency.

[0009] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 Schematic diagram of the flow of the federated learning method based on the disease-specific database described in an embodiment of the present invention.

[0012] Figure 2 Schematic diagram of the structure of a federated learning device based on a disease-specific database described in an embodiment of the present invention.

[0013] Figure 3 Schematic diagram of the structure of a federated learning device based on a disease-specific database described in an embodiment of the present invention.

[0014] In the figure, it is marked as follows: 901, first acquisition module; 902, first processing module; 903, second acquisition module; 904, second processing module; 905, third processing module; 9021, search unit; 9022, first processing unit; 9023, screening unit; 9041, first acquisition unit; 9042, identification unit; 9043, second processing unit; 9044, third processing unit; 9045, fourth processing unit; 9046, fifth processing unit; 90221, first calculation unit; 90222, second calculation unit; 904 31. The sixth processing unit; 90432. The judgment unit; 90433. The seventh processing unit; 90434. The eighth processing unit; 90451. The second acquisition unit; 90452. The ninth processing unit; 90453. The tenth processing unit; 90454. The eleventh processing unit; 90455. The twelfth processing unit; 90456. The thirteenth processing unit; 800. The federated learning device based on the disease-specific database; 801. The processor; 802. The memory; 803. The multimedia component; 804. The I / O interface; 805. The communication component. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0016] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance. Embodiment 1:

[0017] This embodiment provides a federated learning method based on a disease-specific database. It can be understood that a scenario can be laid out in this embodiment, for example: a scenario in which the disease-specific databases of multiple medical institutions need to be federated learned.

[0018] See also Figure 1 , the figure shows that the method includes step S1, step S2, step S3, step S4 and step S5, which are specifically: Step S1: Obtain federated learning request information; Step S2: determining a first node according to the federated learning request information, where the first node is a node that publishes a training task; In this step, after a node that needs to perform federated learning broadcasts a federated learning request message, the verification node screens it to determine the first node.

[0019] It can be understood that the step S2 also includes step S21, step S22 and step S23, which are specifically: Step S21, searching the blockchain for the reputation value corresponding to the first node to obtain a search result; In this step, the reputation value of each node is stored in the blockchain as traceable evidence. Therefore, the reputation value of each node can be queried in the blockchain to determine whether the node has been used before.

[0020] Step S22: Calculate the reputation value of the first node according to the search result to obtain reputation value information; It can be understood that the step S22 also includes step S221 and step S222, which are specifically: Step S221: When the search result shows that the first node does not have a corresponding reputation value, the reputation value of the first node is calculated to obtain first reputation value information; In this step, if the reputation value corresponding to the first node cannot be found in the blockchain, it means that this node is used for the first time, and its reputation value is calculated as follows: ; In the above formula, v represents the reputation value of the node after this training task, that is, the first reputation value information, T, F, and U respectively represent the number of credible behaviors, uncredible behaviors, and uncertain behaviors recorded in the blockchain in all reputation opinions of the node in this round of tasks. They represent the influencing parameters of credible behavior, uncredible behavior, and uncertain behavior in the result respectively.

[0021] Step S222: When the search result shows that the first node has a corresponding reputation value, the reputation value of the first node is recalculated according to the reputation value obtained by the query to obtain second reputation value information.

[0022] In this step, when the reputation value corresponding to the first node is queried in the blockchain, it indicates that the node is not used for the first time. Considering the complexity of the node and the network, as time goes by, the credibility of the node's previous behavior gradually decreases, and the referenceability of the corresponding direct reputation value also gradually decreases. Therefore, the reputation value of the first node needs to be recalculated and determined, and the specific calculation process is as follows: ; In the above formula, Indicates that node i is the task publishing node, i.e., the first node. Represents the second reputation value information, T (t i ) is the introduced time decay function, is the reputation value found in the blockchain, Indicates the decay parameter of time freshness, which is used to adjust the influence of time on the final reputation value. Indicates the record time of the reputation value of the queried node i, Represents all reputation values ​​of node i queried in the blockchain as the task publishing node.

[0023] Step S23: Filter the first node according to the reputation value information to obtain the filtered first node.

[0024] In this step, if the reputation value information is lower than the set threshold, the node is screened out, and the first node corresponding to the value greater than the threshold is screened as the node for publishing the training task.

[0025] Step S3: acquiring training task information according to the first node, wherein the training task information includes a training task sent by the first node to at least one second node, where the second node is a node that executes the training task locally; In this step, a second node corresponds to a local end of a medical institution, and the training task information is sent to multiple second nodes through the first node to realize model training at the local end.

[0026] Step S4: constructing a disease-specific database on the local side based on the training task information as training data to train the local model, thereby obtaining a trained local model; It can be understood that the step S4 also includes step S41, step S42, step S43, step S44, step S45 and step S46, which are specifically: Step S41, obtaining electronic medical record information; In this step, since the electronic medical record includes all document records generated during the medical process, including applications, reports, execution records, informed consent, medical and nursing documents and other information, the electronic medical record is an integration of medical information. Building a disease-specific database based on the electronic medical record can effectively improve the efficiency of building a disease-specific database.

[0027] Step S42: Identify the electronic medical record information to obtain text information; In this step, identifying the electronic medical record information to obtain corresponding text information is a technical solution well known to those skilled in the art, so it will not be described in detail here.

[0028] Step S43, performing term normalization on the text information to obtain term normalized text information; In this step, since there are large differences between medical professional terminology and clinical terminology in electronic medical records, it is very necessary to normalize the terminology written by doctors in electronic medical records.

[0029] It can be understood that the step S43 also includes step S431, step S432, step S433 and step S434, which are specifically: Step S431, sending the text information to the medical entity extraction model to determine the entity information and entity category corresponding to the text information; In this step, establishing a medical entity extraction model is a technical solution well known to those skilled in the art, so it will not be described here in detail. By sending the text information to the medical entity extraction model, the entity information and entity category corresponding to the text information can be determined.

[0030] Step S432: determine whether the entity information matches the vocabulary included in the standard vocabulary corresponding to the entity category, and obtain a matching result; In this step, the standard word library can be a standard word disease library, symptom library, drug library, ICD-9, ICD-10, etc. formed by national open medical standard terms.

[0031] Step S433, determining the target standard word corresponding to the entity information according to the matching result; In this step, if the match is successful in this step, it means that the vocabulary is already a medical term and does not need to be normalized; if the match fails, the entity information and the standard vocabulary are subjected to TF-IDF calculation, a first calculation result, and the vocabulary corresponding to the first calculation result greater than the preset first threshold is screened to obtain the first candidate standard word information; the entity information and the standard vocabulary are subjected to Word2vec similarity calculation, a second calculation result is obtained, and the vocabulary corresponding to the second calculation result greater than the preset second threshold is screened to obtain the second candidate standard word information; the overlapping standard words in the first candidate standard word information and the second candidate standard word information are screened out to obtain the third candidate standard word; the entity information, the text information in the electronic medical record, the entity category and the third candidate standard word are input into the trained binary classification model to obtain the standard word with the highest probability, namely the target standard word.

[0032] Step S434: Use the target standard word to replace the text information to obtain the term-normalized text information.

[0033] Step S44, obtaining a modified electronic medical record according to the normalized text information of the terminology and a preset medical record template; In this step, since different specialized departments have different habits in recording electronic medical records, it is necessary to unify the medical record templates. The electronic medical records are modified using the preset medical record templates to obtain modified electronic medical records. The modified electronic medical records achieve unified standards for templates and terminology.

[0034] Step S45, performing a quality assessment on the modified electronic medical record to obtain an assessment result; In this step, the quality of the electronic medical records is related to the data quality of the model training. Therefore, it is necessary to conduct a quality assessment on the modified electronic medical records to ensure the data quality of the model training.

[0035] It can be understood that the step S45 also includes step S451, step S452, step S453, step S454, step S455 and step S456, which are specifically: Step S451, obtaining preset scoring point information, wherein the scoring point information includes key medical information; In this step, a high-quality electronic medical record should include all important information, namely key medical information, namely scoring points. The quality assessment of the electronic medical record should determine whether the electronic medical record records the key medical information corresponding to the scoring points.

[0036] Step S452: preprocessing the key medical information to obtain preprocessed key medical information; In this step, the key medical information is preprocessed, including labeling it and converting the key medical information into a set of labeled sentences. It should be noted that the labeled labels include C, M, and I, where C indicates that the key medical information can be found in the electronic medical record, M indicates that the key medical information cannot be found in the electronic medical record, and I indicates that the key medical information is incorrectly described in the electronic medical record.

[0037] Step S453, performing calculation according to the preprocessed key medical information to obtain first sentence vector information; In this step, using the word2vec algorithm to convert key medical information into first sentence vector information is a technical solution well known to those skilled in the art.

[0038] Step S454, preprocessing the modified electronic medical record to obtain preprocessed electronic medical record information; In this step, the preprocessing of the electronic medical records includes word segmentation, removal of repeated words, etc.

[0039] Step S455: Calculate according to the preprocessed electronic medical record information to obtain a second sentence vector; In this step, the preprocessed electronic medical record information is converted into the second sentence vector information using the word2vec algorithm, which is a technical solution well known to those skilled in the art. Step S456: Calculate the semantic similarity between the first sentence vector and the second sentence vector, and concatenate the first sentence vector and the second sentence vector with the highest semantic similarity and send them to the trained convolutional neural network model to obtain an evaluation result.

[0040] In this step, a new text set can be obtained by concatenating the first sentence vector and the second sentence vector with the highest semantic similarity. The convolutional neural network model can be trained with the new text set to obtain a trained convolutional neural network model. It should be noted that before training, preprocessing is required to complete the two sentences in the concatenated text set separately.

[0041] In this embodiment, after the convolutional neural network predicts the label of the cascade text, calculations are performed based on the prediction results to obtain evaluation results, which are specifically: ;

[0042] In the above formula, S is the evaluation result of the electronic medical record, They are the number of C labels, the data of N labels, and the number of M labels.

[0043] Step S46: Screen the modified electronic medical records according to the evaluation results to construct a disease-specific database.

[0044] In this step, the quality of the disease-specific database can be controlled by manually setting the score threshold.

[0045] Step S5: Send the trained local model of each second node to the first node for aggregation to obtain a global model. Embodiment 2:

[0046] like Figure 2 As shown, this embodiment provides a federated learning device based on a disease-specific database, the device comprising a first acquisition module 901, a first processing module 902, a second acquisition module 903, a second processing module 904 and a third processing module 905, wherein specifically: A first acquisition module 901 is used to acquire federated learning request information; A first processing module 902 is used to determine a first node according to the federated learning request information, where the first node is a node that publishes a training task; A second acquisition module 903 is used to acquire training task information according to the first node, where the training task information includes a training task sent by the first node to at least one second node, where the second node is a node that executes the training task locally; The second processing module 904 is used to construct a disease-specific database on the local side based on the training task information as training data to train the local model, thereby obtaining a trained local model; The third processing module 905 is used to send the trained local model of each second node to the first node for aggregation to obtain a global model.

[0047] In a specific implementation of the present disclosure, the first processing module 902 includes a search unit 9021, a first processing unit 9022 and a screening unit 9023, wherein specifically: A search unit 9021 is used to search the reputation value corresponding to the first node in the blockchain to obtain a search result; A first processing unit 9022 is used to calculate the reputation value of the first node according to the search result to obtain reputation value information; The screening unit 9023 is used to screen the first node according to the reputation value information to obtain the screened first node.

[0048] In a specific implementation of the present disclosure, the first processing unit 9022 includes a first calculation unit 90221 and a second calculation unit 90222, wherein specifically: A first calculation unit 90221 is configured to calculate the reputation value of the first node when the search result shows that the first node does not have a corresponding reputation value, and obtain first reputation value information; The second calculation unit 90222 is used to recalculate the reputation value of the first node according to the reputation value obtained by the query to obtain second reputation value information when the search result shows that the first node has a corresponding reputation value.

[0049] In a specific implementation of the present disclosure, the second processing module 904 further includes a first acquisition unit 9041, an identification unit 9042, a second processing unit 9043, a third processing unit 9044, a fourth processing unit 9045 and a fifth processing unit 9046, wherein specifically: A first acquisition unit 9041 is used to acquire electronic medical record information; The identification unit 9042 is used to identify the electronic medical record information to obtain text information; The second processing unit 9043 is used to perform term normalization on the text information to obtain term normalized text information; The third processing unit 9044 is used to obtain a modified electronic medical record according to the text information after the term normalization and a preset medical record template; The fourth processing unit 9045 is used to perform quality assessment on the modified electronic medical record to obtain an assessment result; The fifth processing unit 9046 is used to screen the modified electronic medical records according to the evaluation results to construct a disease-specific database.

[0050] In a specific implementation of the present disclosure, the second processing unit 9043 includes a sixth processing unit 90431, a judgment unit 90432, a seventh processing unit 90433 and an eighth processing unit 90434, wherein specifically: The sixth processing unit 90431 is used to send the text information to the medical entity extraction model to determine the entity information and entity category corresponding to the text information; A judging unit 90432 is used to judge whether the entity information matches the vocabulary included in the standard vocabulary corresponding to the entity category, and obtain a matching result; A seventh processing unit 90433 is used to determine a target standard word corresponding to the entity information according to the matching result; The eighth processing unit 90434 is used to replace the text information with the target standard word to obtain the text information after the term normalization.

[0051] In a specific implementation of the present disclosure, the fourth processing unit 9045 includes a second acquisition unit 90451, a ninth processing unit 90452, a tenth processing unit 90453, an eleventh processing unit 90454, a twelfth processing unit 90455, and a thirteenth processing unit 90456, wherein specifically: The second acquisition unit 90451 is used to acquire preset scoring point information, where the scoring point information includes key medical information; A ninth processing unit 90452 is used to pre-process the key medical information to obtain pre-processed key medical information; A tenth processing unit 90453 is used to perform calculation according to the pre-processed key medical information to obtain first sentence vector information; an eleventh processing unit 90454, configured to pre-process the modified electronic medical record to obtain pre-processed electronic medical record information; A twelfth processing unit 90455 is used to perform calculation according to the preprocessed electronic medical record information to obtain a second sentence vector; The thirteenth processing unit 90456 is used to calculate the semantic similarity between the first sentence vector and the second sentence vector, and cascade the first sentence vector and the second sentence vector with the highest semantic similarity to send them to the trained convolutional neural network model to obtain an evaluation result.

[0052] It should be noted that, regarding the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here. Embodiment 3:

[0053] Corresponding to the above method embodiment, this embodiment also provides a federated learning device based on a special disease database. The federated learning device based on a special disease database described below and the federated learning method based on a special disease database described above can refer to each other.

[0054] Figure 3 FIG. 8 is a block diagram of a federated learning device 800 based on a disease-specific database according to an exemplary embodiment. Figure 3 As shown, the federated learning device 800 based on a disease-specific database may include: a processor 801 and a memory 802. The federated learning device 800 based on a disease-specific database may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0055] The processor 801 is used to control the overall operation of the federated learning device 800 based on the disease-specific database to complete all or part of the steps in the above-mentioned federated learning method based on the disease-specific database. The memory 802 is used to store various types of data to support the operation of the federated learning device 800 based on the disease-specific database, and these data may include, for example, instructions for any application or method operating on the federated learning device 800 based on the disease-specific database, and application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 802 or sent via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be keyboards, mice, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the federated learning device 800 based on the disease database and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include: Wi-Fi module, Bluetooth module, NFC module.

[0056] In an exemplary embodiment, the federated learning device 800 based on a disease-specific database can be implemented by one or more application-specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned federated learning method based on a disease-specific database.

[0057] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned federated learning method based on a disease-specific database are implemented. For example, the computer-readable storage medium may be the above-mentioned memory 802 including program instructions, and the above-mentioned program instructions may be executed by the processor 801 of the federated learning device 800 based on a disease-specific database to complete the above-mentioned federated learning method based on a disease-specific database. Embodiment 4:

[0058] Corresponding to the above method embodiment, a readable storage medium is also provided in this embodiment. The readable storage medium described below and the federated learning method based on a special disease database described above can refer to each other.

[0059] A readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the federated learning method based on a disease-specific database of the above-mentioned method embodiment.

[0060] The readable storage medium may specifically be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or other readable storage medium that can store program codes.

[0061] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0062] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A federated learning method based on a disease-specific database, characterized in that: include: Get federated learning request information; Determine a first node according to the federated learning request information, where the first node is a node that publishes a training task; Acquire training task information according to the first node, where the training task information includes a training task sent by the first node to at least one second node, where the second node is a node that executes the training task locally; Based on the training task information, a disease-specific database is constructed locally as training data to train the local model, thereby obtaining a trained local model; The trained local model of each second node is sent to the first node for aggregation to obtain a global model.

2. The federated learning method based on a disease-specific database according to claim 1, characterized in that: Determining a first node according to the federated learning request information includes: Search the reputation value corresponding to the first node in the blockchain to obtain the search result; Calculate the reputation value of the first node according to the search result to obtain reputation value information; The first node is screened according to the reputation value information to obtain a screened first node.

3. The federated learning method based on a disease-specific database according to claim 2, characterized in that: The reputation value of the first node is calculated according to the search result to obtain reputation value information, including: When the search result is that the first node does not have a corresponding reputation value, the reputation value of the first node is calculated, and the first reputation value information; When the search result shows that the first node has a corresponding reputation value, the reputation value of the first node is recalculated according to the reputation value obtained by the query to obtain second reputation value information.

4. The federated learning method based on a disease-specific database according to claim 1, characterized in that: Based on the training task information, a disease-specific database is constructed locally as training data to train the local model, including: Access to electronic medical record information; Identify the electronic medical record information to obtain text information; Normalizing the text information by terms to obtain text information after term normalization; Obtaining a modified electronic medical record according to the normalized text information of the terminology and a preset medical record template; Performing a quality assessment on the modified electronic medical record to obtain an assessment result; The modified electronic medical records are screened according to the evaluation results to construct a disease-specific database.

5. The federated learning method based on a disease-specific database according to claim 4, characterized in that: The text information is terminologically normalized to obtain terminologically normalized text information, including: Sending the text information to a medical entity extraction model to determine entity information and entity categories corresponding to the text information; Determine whether the entity information matches the vocabulary included in the standard vocabulary corresponding to the entity category, and obtain a matching result; Determine the target standard word corresponding to the entity information according to the matching result; The text information is replaced with the target standard word to obtain the term-normalized text information.

6. A federated learning device based on a disease-specific database, characterized in that: include: A first acquisition module is used to obtain federated learning request information; A first processing module, configured to determine a first node according to the federated learning request information, where the first node is a node that publishes a training task; A second acquisition module is used to acquire training task information according to the first node, wherein the training task information includes a training task sent by the first node to at least one second node, and the second node is a node that executes the training task locally; A second processing module is used to construct a disease-specific database on the local side based on the training task information as training data to train the local model, thereby obtaining a trained local model; The third processing module is used to send the trained local model of each second node to the first node for aggregation to obtain a global model.

7. The federated learning device based on a disease-specific database according to claim 6, characterized in that: The first processing module comprises: A search unit, used to search the reputation value corresponding to the first node in the blockchain to obtain a search result; A first processing unit, configured to calculate the reputation value of the first node according to the search result to obtain reputation value information; The screening unit is used to screen the first node according to the reputation value information to obtain the screened first node.

8. The federated learning device based on a disease-specific database according to claim 7, characterized in that: The first processing unit comprises: A first calculation unit, configured to calculate the reputation value of the first node when the search result shows that there is no corresponding reputation value for the first node, and obtain first reputation value information; The second calculation unit is used to recalculate the reputation value of the first node according to the reputation value obtained by the query to obtain second reputation value information when the search result shows that the first node has a corresponding reputation value.

9. The federated learning device based on a disease-specific database according to claim 6, characterized in that: The second processing module comprises: A first acquisition unit, used to acquire electronic medical record information; An identification unit, used to identify the electronic medical record information to obtain text information; A second processing unit, configured to perform term normalization on the text information to obtain term normalized text information; A third processing unit, configured to obtain a modified electronic medical record according to the term normalized text information and a preset medical record template; A fourth processing unit is used to perform a quality assessment on the modified electronic medical record to obtain an assessment result; The fifth processing unit is used to screen the modified electronic medical records according to the evaluation results to build a disease-specific database.

10. The federated learning device based on a disease-specific database according to claim 9, characterized in that: The second processing unit comprises: a sixth processing unit, configured to send the text information to a medical entity extraction model to determine entity information and entity categories corresponding to the text information; A judging unit, used to judge whether the entity information matches the vocabulary included in the standard vocabulary corresponding to the entity category, and obtain a matching result; A seventh processing unit, configured to determine a target standard word corresponding to the entity information according to the matching result; The eighth processing unit is used to replace the text information with the target standard word to obtain the text information after the term normalization.

Images