A data processing system of a medical model federated learning framework

Abstract

The embodiment of the present application relates to a kind of medical model federation learning framework data processing system, the system includes: center network and multiple one, two class agency network;Center network includes center server and pedestal model, center server is connected with pedestal model and each first, second agency server respectively;One class agency network includes first agency server, first sample set and first local model, first agency server is connected with first sample set, first local model respectively;Two class agency network includes second agency server, second sample set and second local model, second agency server is connected with second sample set, second local model respectively, still with another one or more first, second agency server interconnection.The system of the present application can expand training sample range, can provide safe transmission guarantee for the model parameter of longitudinal transmission, can provide safe transmission guarantee for shared information when sharing training data in horizontal agency, provide privacy protection for non-shared information.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to a data processing system of a medical model federated learning framework. BACKGROUND

[0002] Artificial Intelligence (AI) models applied in the medical field are collectively referred to as medical models, which are commonly used in the medical field, such as text object processing models for medical document classification and identification, visual object processing models for medical image organization segmentation or lesion identification, and feature prediction models for disease risk prediction based on medical features. Before using various medical models, they need to be fully trained, and insufficient training data will reduce the generalization ability of the model. At present, a large amount of medical data is mainly distributed in various medical institutions (such as hospitals, medical teaching / research institutions, etc.), and if the traditional centralized training mechanism is used, it is necessary to first collect and store a large amount of data of all institutions, and then configure large computing resources to train the model centrally, which undoubtedly brings great challenges in storage resource configuration, computing resource configuration, privacy data protection, etc.

[0003] Federated learning is a technical framework for distributed model training. Based on this technical framework, it not only reduces the risk of privacy data leakage, but also effectively prevents the center server from causing storage and / or computing resource overdraft risks. There are two types of subdivided frameworks of this technical framework: vertical federated learning (VFL) framework and horizontal federated learning (HFL) framework. By comparing the two subdivided frameworks, it can be found that the same points of the two are: setting a center server as a center node in the center network, and setting an institution server as a distributed node in the institution network of each medical institution. Each distributed node trains a medical model based on local institution sample data, and the center server only aggregates all node model parameters and publishes the new version of the model parameters to the entire network. The difference between the two is that: 1) in the vertical federated learning framework, the sample data of each distributed node has all feature types of the model input features by default, so it only needs to be trained based on local sample data. Each distributed node only transmits model parameters to the center node, and there is no cross-node data transmission between nodes; 2) in the horizontal federated learning framework, the sample data of each distributed node does not have complete model input feature types by default, and the sample features of different distributed nodes can complement each other. Therefore, each distributed node not only transmits model parameters to the center node, but also transmits training data between nodes.

[0004] Most of the current medical model distributed training frameworks are based on the vertical federated learning framework, which requires that the sample features of all distributed nodes in the framework of the institution are rich and complete. But in reality, only high-level hospitals (such as three A-level hospitals) can meet this data requirement. That is to say, the medical model distributed training framework based on the vertical federated learning mechanism can only set up distributed nodes in high-level hospitals, which limits the node access range (low-level hospitals cannot access) and the training sample range (the medical data of low-level hospitals cannot be included in the training sample set).

[0005] To solve this problem, we construct an integrated learning framework by adding a horizontal federated learning framework to the original vertical federated learning framework. The integrated learning framework takes the network of high-level medical institutions as a class of institution networks, and maintains the vertical federated learning framework of all class institution networks. In addition, the network of low-level hospitals is taken as a second class of institution networks, and a horizontal federated learning framework is constructed to access the second class of institution networks and build a training data transmission channel between the second class of institution networks. Based on this integrated learning framework, the node access range and the training sample range can be further expanded. But at the same time, it also increases the management complexity of the whole system data transmission: 1) to ensure the security of model parameter transmission between each class and the center network (the network where the center server is located); 2) to ensure the transmission security of shared data and the non-leakage of non-shared data when sharing training data between horizontal institutions. This is the technical problem that the present invention needs to solve. SUMMARY

[0006] The medical model federated learning framework data processing system provided by the present application can expand the node access range and training sample range through the added horizontal federated learning framework, improve the transmission security of the institution model parameters through the homomorphic encryption mechanism, and ensure the transmission security of the shared information of the horizontal institutions and improve the privacy protection of the non-shared information through the PSI method realized by the OT protocol.

[0007] To achieve the above-mentioned purpose, the embodiment of the present application provides a medical model federated learning framework data processing system, which comprises a center network, a plurality of first-class institution networks and a plurality of second-class institution networks.

[0008] The center network corresponds to a data center; the center network comprises a center server and a base model; the center server is connected with the base model; the center server is also connected with first and second institution servers of each first and second institution network respectively;

[0009] The base model is a preset medical model;

[0010] The center server is configured to establish a secure data transmission channel with each of the first and second institution servers according to a preset key agreement protocol, which is referred to as a corresponding first channel;

[0011] The center server is also configured to periodically synchronize a model feature set of the medical model and a public parameter set required by an oblivious transfer protocol to all institution servers through all the first channels;

[0012] The center server is also configured to periodically take the latest version of model parameters of the base model as a corresponding previous version of base parameter set, and generate a pair of public and private keys for homomorphic encryption and decryption calculation based on a preset parameter aggregation mode, which is referred to as a corresponding public and private key And distribute a model training notification carrying the public key And the previous version of base parameter set to all institution servers through all the first channels; the parameter aggregation mode comprises an unweighted mode and a weighted mode;

[0013] The center server is also configured to, after receiving all institution parameter set ciphertexts returned by all institution servers through all the first channels, perform new version parameter aggregation processing based on the parameter aggregation mode, the private key The previous version of base parameter set and all the institution parameter set ciphertexts to obtain a corresponding new version of base parameter set; and save the new version of base parameter set as the latest version of model parameters of the base model;

[0014] The first institution network corresponds to a medical structure; the first institution network comprises the first institution server, a first sample set and a first local model; the first institution server is connected with the first sample set and the first local model respectively;

[0015] The first local model is consistent with the base model in model structure but different in model parameters;

[0016] The first institution server is configured to store the received model feature set, public parameter set and model training notification;

[0017] The first institution server is further configured to, upon receiving each new model training notification, perform a round of local training on the first local model based on the model training notification and the first sample set to obtain a corresponding institution parameter set ciphertext and send the institution parameter set ciphertext to the center server.

[0018] The second institution network corresponds to a medical structure; the second institution network comprises a second institution server, a second sample set and a second local model; the second institution server is connected with the second sample set and the second local model respectively; each second institution server is further interconnected with one or more first and second institution servers;

[0019] The second local model has the same model structure as the base model but different model parameters;

[0020] The second institution server is configured to store the received model feature set, the public parameter set and the model training notification;

[0021] The second institution server is further configured to, upon receiving each new model training notification, extract a corresponding public key from the current model training notification and the previous version base parameter set; perform parameter initialization on the second local model based on the previous version base parameter set; perform missing feature checking on the second local model based on the model feature set and the second sample set to obtain a corresponding missing feature set X; perform sample feature completion processing on the second sample set through a round of interaction with other institution servers connected therewith based on the public parameter set and the missing feature set X to obtain a corresponding third sample set; and perform a round of local training on the second local model based on the third sample set to obtain a corresponding institution parameter set ciphertext and send the institution parameter set ciphertext to the center server.

[0022] Preferably, the model feature set comprises a plurality of model input / output features, each of which corresponds to a type of medical feature information.

[0023] The oblivious transfer protocol is based on the discrete logarithm problem; the public parameter set of the oblivious transfer protocol comprises cyclic group parameters G, generator parameters g of a cyclic group, order parameters q of the cyclic group, a hash function H ID for key derivation KEY ; the hash function H ID takes a patient's social identity as input and outputs a group element in the cyclic group G; the hash function H KEYThe input of the function is a cyclic group element power operation result, and the output is a derived key with a specified key length; the social identity includes at least an ID card identity;

[0024] The first sample set includes a plurality of first sample records; each first sample record is a historical medical record of a patient, including the social identity of the current patient and all the medical feature information of the current patient in the current medical structure; the corresponding patients of all the first sample records are not repeated; the types of the medical feature information include at least medical record text, physical examination report text, medical examination report text, medical examination report text, medical examination image, medical examination audio, treatment plan text, and diagnosis record text; all the medical feature information of the first sample records covers all the model input / output features of the model feature set;

[0025] The second sample set includes a plurality of second sample records; each second sample record is a historical medical record of a patient, including the social identity of the current patient and all the medical feature information of the current patient in the current medical structure; the corresponding patients of all the second sample records are not repeated; all the medical feature information of the second sample records partially or entirely covers all the model input / output features of the model feature set; the medical feature information types of all the second sample records in the second sample set are aligned by default, and there is no difference in the type range of the medical feature information between any two second sample records;

[0026] The third sample set includes a plurality of third sample records; each third sample record is a medical record of a patient, including the social identity of the current patient and all the medical feature information of the current patient in one or more medical structures; the corresponding patients of all the third sample records are not repeated; all the medical feature information of the third sample records covers all the model input / output features of the model feature set;

[0027] When the missing feature set X is not empty, one or more missing features x j are included, 1≤index j≤N X , N X is the total number of missing features in the current missing feature set; each missing feature x j corresponds to one model input / output feature in the model feature set.

[0028] Preferably, the key agreement protocol includes at least DH protocol, ECDH protocol, IKEv2 protocol, and TLS-PSK protocol;

[0029] The two servers on both sides of each first channel have a pair of shared keys;

[0030] The transmission mechanism of each first channel is that the center server or the first and second institution servers as the current channel data sender encrypts the original plaintext based on the shared key and sends the obtained ciphertext to the data receiver on the other side of the current channel, and the first and second institution servers or the center server as the current channel data receiver decrypts the ciphertext based on the shared key to obtain the corresponding original plaintext.

[0031] Preferably, the center server is specifically used to generate a pair of public and private keys for homomorphic encryption and decryption calculation for the previous version base parameter set based on the preset parameter aggregation mode, denoted as the corresponding public and private keys ;

[0032] Identify the parameter aggregation mode;

[0033] If the parameter aggregation mode is a non-weighted mode, the public and private key pair generation process is performed based on a preset semi-homomorphic algorithm to obtain the corresponding public key and the private key The semi-homomorphic algorithm at least includes a Paillier algorithm;

[0034] If the parameter aggregation mode is a weighted mode, the public and private key pair generation process is performed based on a preset full-homomorphic algorithm to obtain the corresponding public key and the private key The full-homomorphic algorithm at least supports BFV algorithm, BGV algorithm, CKKS algorithm, FHEW algorithm, and TFHE algorithm.

[0035] Preferably, the center server is specifically used to perform new version parameter aggregation processing on the previous version base parameter set and all institution parameter set ciphertexts based on the parameter aggregation mode and the private key to obtain the corresponding new version base parameter set;

[0036] The total number of institution parameter set ciphertexts is counted to obtain the corresponding total number N; and each institution parameter set ciphertext is denoted as the corresponding ciphertext EW i , 1≤index i≤N; the previous version base parameter set is denoted as the corresponding parameter set W old ; and the new version base parameter set is denoted as the corresponding parameter set W new ;

[0037] And identify the parameter aggregation mode;

[0038] If the parameter aggregation mode is a non-weighted mode, the public and private key pair generation process is performed based on a preset semi-homomorphic algorithm to obtain the corresponding public key corresponding new version differential parameter set ciphertext EW new ;

[0039] If the parameter aggregation mode is a non-weighted mode, the corresponding new version differential parameter set ciphertext EW is generated in a manner that new ; i is the preset weight coefficient corresponding to the i-th ciphertext EW i ;

[0040] and the new version differential parameter set ciphertext EW new is homomorphically decrypted based on the private key new to obtain the corresponding new version differential parameter set AW new ; is a preset homomorphic decryption function used to decrypt the new version differential parameter set ciphertext EW old using the private key new ; and the corresponding parameter set W new is calculated based on the parameter set W new and the new version differential parameter set AW old , W new = W now + AW.

[0041] Preferably, the first institution server is specifically configured to, when the first local model is trained based on the model training notification and the first sample set to obtain the corresponding institution parameter set ciphertext EW

[0042] and the center server sends back:

[0043] extract the corresponding public key and the previous version base parameter set from the model training notification;

[0044] and initialize the parameters of the first local model based on the previous version base parameter set;

[0045] and train the first local model that has completed initialization based on the first sample set for one round;

[0046] and record the current model parameter set of the first local model as parameter set W now at the end of this round of training, record the previous version base parameter set as parameter set W old , and calculate the latest differential parameter set AW = W now -W old based on the parameter set W X , W ;

[0047] and based on the public key encrypt the difference parameter set △W to obtain the corresponding agency parameter set ciphertext; for a preset homomorphic encryption function, for using the public key encrypt the latest difference parameter set △W;

[0048] and send the current agency parameter set ciphertext to the center server through the corresponding first channel.

[0049] Preferably, the second agency server is specifically configured to, when the missing feature set X is obtained based on the model feature full set and the second sample set, identify all medical features covered by the second sample set to obtain a corresponding medical feature set; identify the feature intersection of the medical feature set and the model feature full set to obtain a corresponding current feature intersection; and take the features in the model feature full set other than the current feature intersection as the corresponding missing feature set X.

[0050] Preferably, the second agency server is specifically configured to, when the third sample set is obtained by one round of interaction according to the public parameter set and the missing feature set X with other agency servers connected with itself and sample feature filling processing of the second sample set through this round of interaction:

[0051] perform one round of traversal on all second sample records of the second sample set; and in the traversal process, take the second sample record currently traversed as a corresponding current sample record; and in the current sample record, add N X medical feature information corresponding to the missing feature x j of the missing feature set X one by one and preset as null information to obtain a corresponding third sample record; and when the round of traversal ends, all third sample records obtained by the round of traversal form an initialized third sample set;

[0052] and perform one round of traversal on all first and second agency servers connected with itself; and in the traversal process, take the first or second agency server currently traversed as a current sender server; and based on the privacy set intersection method implemented by the inadvertent transmission protocol, obtain sample missing features from the sender server according to the public parameter set and the missing feature set X to fill the third sample set; and when the round of traversal ends, output the latest third sample set as the processing result of the sample feature filling processing.

[0053] Furthermore, the second agency server is specifically used to fill the third sample set with the sample missing features obtained from the sending server according to the public parameter set and the missing feature set X, in the privacy set intersection method implemented based on the unintentional transmission protocol.

[0054] Send the interaction request carrying the missing feature set X to the sending server;

[0055] The total number N is obtained by counting the total number of records in the third sample set. U ; and input each of the social identity identifiers in the third sample set into the hash function H. ID Perform the calculation and use the result as the corresponding identifier element h. u =H ID (ID u ), 1 ≤ index u ≤ N U ID u h is the u-th social identity identifier in the third sample set. u ∈G;

[0056] And based on the order parameter q, the corresponding remainder ring parameter Z is set. q {0,1,2,…,q-1};

[0057] And the remainder ring parameter Z q Perform N U Each random number is selected and used as a corresponding patient private key S. u S u ∈Z q Based on the generated meta-parameter g and each of the patient's private keys S u and its corresponding identifier element h u Generate a corresponding patient public key C u , C u ∈G;

[0058] After receiving the sender's public key H sent back by the sender's server, N will... U The patient's public key C u The patient public key set {C u} Send to the sending server; and receive the ciphertext matrix {ED} sent back by the sending server. u,v}; the ciphertext matrix {ED u,v} by N U ×N V ciphertext data ED u,v composition;

[0059] And based on the hash function HKEY , the sender public key H and each of the patient private key S u generate a corresponding patient shared key K u ,

[0060] and based on each of the patient shared key K u decrypt the corresponding N V ciphertext data ED u,v get the corresponding N V plaintext data DED u,v , DED u,v = f DE (ED u,v , K u,v ), f DE is a preset decryption function of symmetric encryption and decryption algorithm;

[0061] and a round of traversal is performed on N U × N V plaintext data DED u,v ; and in the current round of traversal, the plaintext data DED u,v currently traversed is taken as the corresponding current plaintext; and according to a preset splicing rule of patient missing information set and check code, the corresponding patient missing information set D u,v and check code MAC u,v are extracted from the current plaintext; and based on a preset check code algorithm, the patient missing information set D u,v is subjected to check code calculation to obtain the corresponding current check code MAC now ; and when the current check code MAC now matches the MAC u,v , the current patient missing information set D u,v is taken as the patient missing information set D u corresponding to the current plaintext, and based on the current patient missing information set D u , N X medical feature information corresponding to the missing feature set X in the u-th third sample record of the third sample set is set.

[0062] Further, the sender server, after receiving one of the interactive requests, takes the second institution server sending the current interactive request as the current receiving server;

[0063] and extracts the corresponding missing feature set X from the current interactive request;

[0064] and takes the local first or second sample set as the corresponding current sample set;

[0065] and the total number N of records of the current sample set is counted V ;

[0066] and each of the social identity identifiers of the current sample set is brought into the hash function H ID is calculated and the calculation result is taken as a corresponding identifier element h v = H ID (ID v ), 1≤index v≤N V , ID v is the vth social identity identifier of the current sample set, h v ∈G;

[0067] and the corresponding remainder ring parameter Z q {0,1,2,…,q-1} is set based on the order parameter q

[0068] and a random number is selected based on the remainder ring parameter Z q and the selected random number is taken as a corresponding sender private key S, S∈Z q ; and a corresponding sender public key H is generated for the sender private key S based on the generator parameter g, H=g S , H∈G;

[0069] and the sender public key H is sent back to the receiver server; and after sending the sender public key H, the patient public key set {C u} sent by the receiver server is received; the patient public key set {C u} is composed of N U patient public keys C u ;

[0070] and a corresponding patient shared key K KEY is generated based on the hash function H u , the patient public key set {C v}, the sender private key S and all the identifier elements h u,v ,

[0071] and the medical feature information corresponding to each of the missing features x j of the missing feature set X in the first or second sample record corresponding to the vth social identity identifier of the current sample set is extracted as a corresponding missing feature information d v,j , if the sample record of the current sample set also does not have the medical feature information corresponding to a missing feature x j , the corresponding missing feature information d v,jis empty; and each of the social identities is identified by N X corresponding to the missing feature information d v,j of the patient v {d v,j};

[0072] and each of the patient missing information sets D v {d v,j} is subjected to a check code calculation based on the preset check code algorithm to obtain a corresponding check code MAC v ;

[0073] and each of the patient missing information sets D v {d v,j} and the corresponding check code MAC v composes a corresponding patient data DM v ;

[0074] and the corresponding patient data DM u,v is encrypted based on each of the patient shared keys K v to obtain the corresponding ciphertext data ED u,v , ED u,v = f EN (D v , K u,v ), f EN is a preset encryption function of a symmetric encryption and decryption algorithm, and f EN and f DE are inverse operation functions of each other;

[0075] and the N U × N V ciphertext data ED u,v composes a corresponding ciphertext matrix {ED u,v} to the receiving server.

[0076] Preferably, the second institution server is specifically configured to, when the second local model based on the third sample set is subjected to one round of local training to obtain the corresponding institution parameter set ciphertext and the center server is fed back:

[0077] based on the third sample set, the second local model that has completed initialization is subjected to one round of training;

[0078] and at the end of the round of training, the current model parameter set of the second local model is recorded as the parameter set W' now , the previous version of the base parameter set is recorded as the parameter set W’ old , and the parameter sets W' now , W’ oldcalculating a latest difference parameter set △W' = W' now -W' old ;

[0079] and encrypting the difference parameter set △W' based on the public key to obtain the corresponding institution parameter set ciphertext; is a preset homomorphic encryption function, and is used to encrypt the latest difference parameter set △W' using the public key ;

[0080] and sending the current institution parameter set ciphertext to the center server through the corresponding first channel.

[0081] The embodiment of the present application provides a kind of medical model federated learning framework data processing system, the system includes: center network, multiple one type of organization network, multiple two type of organization network;Center network is arranged in data center, one type of organization network is arranged in high-grade medical institution, two type of organization network is arranged in low-grade hospital institution;Center network and the federated learning framework of all one type of organization network belongs to vertical federated learning framework, and one type of organization server under vertical framework has no interconnection relationship with each other, only to the center server connection;Center network and the federated learning framework of all two type of organization network belongs to horizontal federated learning framework, and two type of organization server under horizontal framework can be interconnected with the service of other one type or two type of organization.The center server of center network and each organization server are carried out data transmission through independent secure data transmission channel, and center server is used to periodically synchronize the model feature set of medical model and the public parameter set required by OT protocol to all organization servers, also be used to periodically distribute the latest version of model parameter of base model to all organization servers to start a round of model training task, also be used to aggregate and decrypt the organization parameter set ciphertext fed back by all organization servers in homomorphic decryption mode to obtain the latest version of model parameter of base model.The server of one type of organization is trained in each round of model training task Local model according to the sample set of the organization, and the organization model parameter obtained by training is encrypted based on homomorphic encryption mode, and the organization parameter set ciphertext obtained is sent to the center server.The server of two type of organization is first obtained from the other organization servers connected horizontally based on the PSI method realized by OT protocol in each round of model training task Supplemental features to improve the sample set of the organization, then the local model is trained based on the improved sample set, then the organization model parameter obtained by training is encrypted based on homomorphic encryption mode, and the organization parameter set ciphertext obtained is sent to the center server.The system of the embodiment of the present application expands the node access range and expands the training sample range on the one hand through the added horizontal federated learning framework;On the other hand, the transmission security of organization model parameter is improved through homomorphic encryption mechanism;Thirdly, through the PSI method realized by OT protocol, the transmission security of shared information of horizontal organization is guaranteed, and the privacy protection intensity of non-shared information is improved. BRIEF DESCRIPTION OF DRAWINGS

[0082] Figure 1 A module structure diagram of the medical model federated learning framework data processing system provided by the embodiment of the present application is provided.

[0083] Figure 2 The interaction schematic diagram of receiver server and sender server provided by the embodiment of the present application. DETAILED DESCRIPTION

[0084] In order to make the objects, technical solutions and advantages of the present application clearer, the following further describes the present application in detail with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0085] The data processing system of the medical model federated learning framework provided by the embodiments of the present application comprises a center network 1, a plurality of first-institution networks 2 and a plurality of second-institution networks 3. Figure 1 The module structure diagram of the data processing system of the medical model federated learning framework provided by the embodiments of the present application is shown in the figure, which mainly comprises a center network 1, a plurality of first-institution networks 2 and a plurality of second-institution networks 3.

[0086] (I) Center network 1:

[0087] The center network 1 of the embodiments of the present application corresponds to a data center.

[0088] The center network 1 comprises a center server 11 and a base model 12; the center server 11 is connected with the base model 12; the center server 11 is also connected with a first-institution server 21 of each first-institution network 2 and a second-institution server 31 of each second-institution network 3 respectively.

[0089] The center server 11 is configured to establish a secure data transmission channel with each first-institution server 21 and second-institution server 31 respectively according to a preset key agreement protocol, which is referred to as a corresponding first channel.

[0090] Here, the key agreement protocol of the embodiments of the present application at least comprises a DH protocol, an ECDH protocol, an IKEv2 protocol and a TLS-PSK protocol.

[0091] The feature of the secure data transmission channel, i.e. the first channel, constructed based on the key agreement protocol is that there is a pair of same shared keys on the two servers on both sides of each channel. The transmission mechanism of the first channel is that the center server 11, the first-institution server 21 or the second-institution server 31 as the data sender of the current channel encrypts the original plaintext based on the shared key and sends the obtained ciphertext to the data receiver on the other side of the current channel, and the first-institution server 21, the second-institution server 31 or the center server 11 as the data receiver of the current channel decrypts the ciphertext based on the shared key to obtain the corresponding original plaintext.

[0092] The center server 11 is also configured to periodically synchronize the model feature set of the medical model and the public parameter set required by the oblivious transfer protocol to all institution servers through all first channels.

[0093] Here, the model feature set of the embodiment of the application comprises a plurality of model input / output features, each of which corresponds to a type of medical feature information.

[0094] The OT protocol of the embodiment of the application is an OT protocol realized based on a discrete logarithm problem; the public parameter set of the OT protocol comprises a cyclic group parameter G of the discrete logarithm problem, a generator parameter g of the cyclic group, a parameter q of the order of the cyclic group, a hash function H ID for key derivation KEY ; wherein the function input of the hash function H ID is a social identity of a patient, and the output is a group element in a cyclic group G; the input of the hash function H KEY is a power operation result of a cyclic group element, and the output is a derived key of a specified key length; the social identity of the embodiment of the application at least comprises an identity card identification.

[0095] The center server 11 is also configured to periodically take the latest version of the model parameters of the base model 12 as the corresponding previous version of the base parameters set; and generate a pair of public and private keys for homomorphic encryption and decryption calculation based on a preset parameter aggregation mode for the previous version of the base parameters set, denoted as a corresponding public key and a private key , and distribute the model training notification carrying the public key and the previous version of the base parameters set to all the institution servers through all the first channels.

[0096] Here, the parameter aggregation mode of the embodiment of the application comprises an unweighted mode and a weighted mode.

[0097] The center server 11 is also configured to, after receiving the ciphertexts of the institution parameters sets returned by all the institution servers through all the first channels, perform new version parameter aggregation processing based on the parameter aggregation mode, the private key , the previous version of the base parameters set and the ciphertexts of all the institution parameters sets to obtain the corresponding new version of the base parameters set; and take the new version of the base parameters set as the latest version of the model parameters of the base model 12 and save it.

[0098] The base model 12 of the embodiment of the application is a type of preset medical model.

[0099] In one specific implementation manner of the embodiment of the application, the center server 11 is specifically configured to, when generating a pair of public and private keys for homomorphic encryption and decryption calculation based on a preset parameter aggregation mode for the previous version of the base parameters set, denoted as a corresponding public key and a private key :

[0100] Step A1, identify the parameter aggregation mode;

[0101] Step A2, if the parameter aggregation mode is a non-weighted mode, a public key and a private key are generated based on a preset semi-homomorphic algorithm

[0102] The semi-homomorphic algorithm at least includes a Paillier algorithm.

[0103] Step A3, if the parameter aggregation mode is a weighted mode, a public key and a private key are generated based on a preset full-homomorphic algorithm

[0104] The full-homomorphic algorithm at least supports a BFV algorithm, a BGV algorithm, a CKKS algorithm, a FHEW algorithm and a TFHE algorithm.

[0105] In another specific implementation of the embodiment of the application, the center server 11 is specifically configured to perform new version base parameter set aggregation processing on the basis of the parameter aggregation mode, the private key

[0106] Step B1, the total number of the institution parameter set ciphertexts is counted to obtain a corresponding total number N; and each institution parameter set ciphertext is recorded as a corresponding ciphertext EW i , 1≤index i≤N; the previous version base parameter set is recorded as a corresponding parameter set W old ; and the new version base parameter set is recorded as a corresponding parameter set W new .

[0107] Step B2, the parameter aggregation mode is identified; if the parameter aggregation mode is a non-weighted mode, a corresponding new version differential parameter set ciphertext EW new is generated in the manner of if the parameter aggregation mode is a non-weighted mode, a corresponding new version differential parameter set ciphertext EW new is generated in the manner of .

[0108] Wherein, α i is a preset weight coefficient corresponding to the i-th ciphertext EW i .

[0109] Step B3, the new version differential parameter set ciphertext EW new is homomorphically decrypted based on the private key to obtain a corresponding new version differential parameter set △W new .

[0110] Wherein, ​​​

[0111] a preset homomorphic decryption function for using a private key to decrypt the ciphertext EW new of the new version differential parameter set;

[0112] Step B4, and calculating the corresponding parameter set W old and the new version differential parameter set AW new based on the parameter set W new .

[0113] wherein W new = W old + AW new .

[0114] (ii) a type of institutional network 2:

[0115] The type of institutional network 2 of the embodiment of the present application corresponds to a high-level medical structure, for example, a three-A-level hospital.

[0116] The type of institutional network 2 includes a first institutional server 21, a first sample set 22 and a first local model 23; the first institutional server 21 is connected with the first sample set 22 and the first local model 23 respectively.

[0117] The first institutional server 21 is configured to store the received model feature set, the public parameter set and the model training notification.

[0118] The first institutional server 21 is further configured to, when receiving a new model training notification, perform a round of local training on the first local model 23 based on the model training notification and the first sample set 22 to obtain a corresponding institutional parameter set ciphertext and send the institutional parameter set ciphertext to the center server 11.

[0119] The first sample set 22 of the embodiment of the present application includes a plurality of first sample records; each first sample record is a historical medical record of a patient, including the social identity of the current patient and all medical feature information of the current patient in the current medical structure. It should be noted that the corresponding patients of all first sample records are not repeated. It should be further noted that the types of medical feature information of the embodiment of the present application at least include medical record text, physical examination report text, medical examination report text, medical examination report text, medical examination image, medical examination audio, treatment plan text and diagnosis record text. It should be further noted that all medical feature information of the first sample record covers all model input / output features of the model feature set.

[0120] The first local model 23 of the embodiment of the present application is consistent with the model structure of the base model 12 but the model parameters are different.

[0121] In another specific implementation manner of the embodiment of the present application, the first institution server 21 is specifically configured to perform one round of local training on the first local model 23 based on the model training notification and the first sample set 22 to obtain a corresponding institution parameter set ciphertext, and send the institution parameter set ciphertext to the center server 11:

[0122] Step C1, extracting the corresponding public key from the model training notification and the previous version base parameter set;

[0123] Step C2, and performing parameter initialization on the first local model 23 based on the previous version base parameter set;

[0124] Step C3, and performing one round of training on the initialized first local model 23 based on the first sample set 22;

[0125] Step C4, and recording the current model parameter set of the first local model 23 as the parameter set W now at the end of the round of training, recording the previous version base parameter set as the parameter set W old , and calculating the latest differential parameter set AW = W now -W old based on the parameter set W now , W old ;

[0126] Step C5, and encrypting the differential parameter set AW based on the public key to obtain a corresponding institution parameter set ciphertext;

[0127] wherein,

[0128] is a preset homomorphic encryption function, used for encrypting the latest differential parameter set AW using the public key ;

[0129] Step C6, and sending the current institution parameter set ciphertext to the center server 11 through the corresponding first channel.

[0130] (Three) the second institution network 3:

[0131] The second institution network 3 corresponds to a medium and low level medical structure, for example, a second-class hospital, a community hospital, etc.

[0132] The second institution network 3 includes a second institution server 31, a second sample set 32 and a second local model 33; the second institution server 31 is connected with the second sample set 32 and the second local model 33 respectively; each second institution server 31 is further interconnected with another one or more first institution servers 21 and second institution servers 31.

[0133] The second institution server 31 is configured to store the received model feature set, the public parameter set, and the model training notification.

[0134] The second institution server 31 is further configured to extract the corresponding public key from the current model training notification and the previous version base parameter set when receiving a new model training notification. and the previous version base parameter set; perform parameter initialization on the second local model 33 based on the previous version base parameter set; perform missing feature checking on the second sample set 32 based on the model feature set and the second sample set 32 to obtain a corresponding missing feature set X; perform a round of interaction with other institution servers connected to itself based on the public parameter set and the missing feature set X, and perform sample feature completion processing on the second sample set 32 through the round of interaction to obtain a corresponding third sample set; and perform a round of local training on the second local model 33 based on the third sample set to obtain a corresponding institution parameter set ciphertext, and send the institution parameter set ciphertext to the center server 11.

[0135] Here, when the missing feature set X of the embodiment of the application is not empty, the missing feature set X is composed of one or more missing features x j , 1≤index j≤N X , N X is the total number of missing features in the current missing feature set; each missing feature x j corresponds to a model input / output feature pair in the model feature set.

[0136] The second sample set 32 of the embodiment of the application includes a plurality of second sample records; each second sample record is a historical medical record of a patient, including the social identity of the current patient and all medical feature information of the current patient in the current medical structure; the corresponding patients of all second sample records are not repeated; all medical feature information of the second sample record covers part or all of all model input / output features of the model feature set; the types of medical feature information of all second sample records in the second sample set 32 are aligned by default, and there is no difference in the type range of medical feature information between any two second sample records.

[0137] The third sample set of the embodiment of the application includes a plurality of third sample records; each third sample record is a medical record of a patient, including the social identity of the current patient and all medical feature information of the current patient in one or more medical structures; the corresponding patients of all third sample records are not repeated; all medical feature information of the third sample record covers all model input / output features of the model feature set.

[0138] The second local model 33 of the embodiment of the application is consistent with the model structure of the base model 12 but differs in model parameters.

[0139] In another specific implementation manner of the embodiment of the present application, the second institution server 31 is specifically configured to: when the corresponding missing feature set X is obtained based on the model feature complete set and the second sample set 32, identify all medical features covered by the second sample set 32 to obtain a corresponding medical feature set; identify the feature intersection of the medical feature set and the model feature complete set to obtain a corresponding current feature intersection; and set the features in the model feature complete set except the current feature intersection as the corresponding missing feature set X.

[0140] In another specific implementation manner of the embodiment of the present application, the second institution server 31 is specifically configured to: when the corresponding third sample set is obtained by performing a sample feature supplementing processing on the second sample set 32 through a round of interaction with other institution servers connected with itself according to the public parameter set and the missing feature set X:

[0141] Step D1, a round of traversal is performed on all second sample records of the second sample set 32; and in the round of traversal, a currently traversed second sample record is taken as a corresponding current sample record; and in the current sample record, N X medical feature information corresponding to the missing feature x j in the missing feature set X and preset as null information, so as to obtain a corresponding third sample record; and when the round of traversal ends, all third sample records obtained through the round of traversal form an initialized third sample set;

[0142] Step D2, a round of traversal is performed on all first institution servers 21 and second institution servers 31 connected with itself; and in the round of traversal, a currently traversed first institution server 21 or second institution server 31 is taken as a current sender server; and a privacy set intersection method implemented based on the oblivious transfer protocol is used to obtain sample missing features from the sender server according to the public parameter set and the missing feature set X to fill the third sample set; and when the round of traversal ends, the latest third sample set is taken as a processing result of the sample feature supplementing processing.

[0143] In another specific implementation manner of the embodiment of the present application, the second institution server 31 is specifically configured to: when the third sample set is filled with sample missing features obtained from the sender server according to the public parameter set and the missing feature set X based on the privacy set intersection method implemented based on the oblivious transfer protocol:

[0144] Step E1, an interaction request carrying the missing feature set X is sent to the sender server;

[0145] Step E2, the total number of records of the third sample set is counted to obtain a corresponding total number N U ; and each social identity identifier of the third sample set is brought into a hash function H ; and each social identity identifier of the third sample set is brought into a hash function HID Perform the calculation and use the result as the corresponding identifier element h. u ;

[0146] Here, h u =H ID (ID u ),

[0147] Where 1 ≤ index u ≤ N U ID u h is the u-th social identity identifier in the third sample set. u ∈G;

[0148] Step E3, and set the corresponding remainder ring parameter Z based on the order parameter q. q {0,1,2,…,q-1};

[0149] Step E4, and adjust the remainder ring parameter Z. q Perform N U Each random number is selected and used as a corresponding patient private key S. u S u ∈Z q Based on the generator meta-parameter g and each patient's private key S u and its corresponding identifier element h u Generate a corresponding patient public key C u ;

[0150] here, C u ∈G;

[0151] Step E5, and after receiving the sender's public key H from the sender's server, will be handled by N. U Patient public key C u The patient public key set {C u Send to the sending server;

[0152] Step E6, and receive the ciphertext matrix {ED} sent back by the sending server. u,v};

[0153] Wherein, the ciphertext matrix {ED u,v} by N U ×N V ciphertext data ED u,v composition;

[0154] Step E7, and based on the hash function H KEY The sender's public key H and each patient's private key S u Generate a corresponding patient shared key K u ;

[0155] here,

[0156] Step E8, and based on the respective patient shared key K u decrypt the corresponding N V cipher data ED u,v obtain the corresponding N V plain data DED u,v ;

[0157] Here, DED u,v = f DE (ED u,v , K u,v ),

[0158] wherein f DE is a preset decryption function of a symmetric encryption and decryption algorithm;

[0159] Step E9, and a round of traversal is performed on the N U × N V plain data DED u,v , and in the round of traversal, the plain data DED u,v being currently traversed is taken as the corresponding current plain text, and the corresponding patient missing information set D u,v and the check code MAC u,v are extracted from the current plain text according to a preset assembling rule of the patient missing information set and the check code; the patient missing information set D u,v is subjected to check code calculation based on a preset check code algorithm to obtain the corresponding current check code MAC now ; and when the current check code MAC now matches the MAC u,v , the current patient missing information set D u,v is taken as the patient missing information set D u corresponding to the current plain text, and the N X medical feature information corresponding to the missing feature set X in the u-th third sample record of the third sample set is set based on the current patient missing information set D u .

[0160] Here, the preset check code algorithm of the embodiment of the application at least includes SHA series hash algorithms, SM3 hash algorithms, and MD series digital digest algorithms. It should be further noted that when the N U × N V plain data DED u,v is subjected to traversal verification, each index u has at most one matching patient missing information set D u .

[0161] The second institution server 31 performing the above steps E1-E9 is regarded as a receiving server, and the processing flow of the sender server side corresponding to the receiving server is as follows:

[0162] Step F1, after receiving an interaction request, the sender server takes the second institution server 31 sending the current interaction request as the current receiving server;

[0163] Step F2, and extracts the corresponding missing feature set X from the current interaction request;

[0164] Step F3, and take the local first or second sample set 32 as the corresponding current sample set;

[0165] Step F4, and count the total number of records of the current sample set to obtain the corresponding total number N V ;

[0166] Step F5, and bring each social identity identifier of the current sample set into the hash function H ID for calculation and take the calculation result as the corresponding identifier element h v ;

[0167] Here, h v = H ID (ID v ),

[0168] Where 1≤index v≤N V , ID v is the vth social identity identifier of the current sample set, h v ∈G;

[0169] Step F6, and set the corresponding residue ring parameter Z q {0,1,2,…,q-1} based on the order parameter q;

[0170] Step F7, and randomly select a random number for the residue ring parameter Z q and take the selected random number as a corresponding sender private key S, S∈Z q ; and generate a corresponding sender public key H for the sender private key S based on the generator parameter g;

[0171] Here, H = g S , H∈G;

[0172] Step F8, and send the sender public key H back to the receiving server;

[0173] Step F9, and after sending the sender public key H, receive the patient public key set {C u} sent by the receiving server;

[0174] Here, the patient public key set {C u} is composed of N U patient public keys C u ;

[0175] Step F10, and based on the hash function H KEY , the patient public key set {C u}, the sender private key S, and all the identity elements h v , generate the corresponding patient shared key K u,v ;

[0176] Here,

[0177] Step F11, and extract the medical feature information corresponding to each missing feature x j of the missing feature set X in the first or second sample record corresponding to the vth social identity in the current sample set as the corresponding missing feature information d v,j , and if the sample record of the current sample set also does not have medical feature information corresponding to a missing feature x j , the corresponding missing feature information d v,j is set to null; and the N X missing feature information d v,j corresponding to each social identity constitutes the corresponding patient missing information set D v {d v,j};

[0178] Step F12, and based on the preset check code algorithm, perform check code calculation on each patient missing information set D v {d v,j} to obtain the corresponding check code MAC v ;

[0179] Step F13, and the patient data DM v is composed of each patient missing information set D v,j {d v} and the corresponding check code MAC v ;

[0180] Step F14, and based on each patient shared key K u,v encrypt the corresponding patient data DM v to obtain the corresponding ciphertext data ED u,v ;

[0181] Here, ED u,v = f EN (D v ,K u,v ),

[0182] where f ENThe encryption function f is a preset symmetric encryption and decryption algorithm EN The encryption function f is a preset symmetric encryption and decryption algorithm DE The encryption function f is a preset symmetric encryption and decryption algorithm

[0183] Step F15, and the obtained N U ×N V The ciphertext data ED u,v is used to form a corresponding ciphertext matrix {ED u,v} and sent to the receiving server.

[0184] The above-mentioned interaction process between the receiving server and the sending server can be understood with reference to Figure 2 The interaction between the receiving server and the sending server provided by the embodiment of the present application can be understood with reference to the interaction schematic diagram between the receiving server and the sending server.

[0185] In another specific implementation manner of the embodiment of the present application, the second institution server 31 is specifically configured to, when sending the corresponding institution parameter set ciphertext to the center server 11 based on the third sample set for performing one round of local training on the second local model 33 to obtain the corresponding institution parameter set ciphertext:

[0186] Step G1, performing one round of training on the initialized second local model 33 based on the third sample set;

[0187] Step G2, and recording the current model parameter set of the second local model 33 as the parameter set W' now at the end of the round of training, recording the previous version of the base parameter set as the parameter set W' old , and calculating the latest difference parameter set △W' = W' now -W' ol based on the parameter sets W' now and W' old ;

[0188] Step G3, and encrypting the difference parameter set △W' based on the public key to obtain the corresponding institution parameter set ciphertext;

[0189] Here,

[0190] The encryption function f is a preset symmetric encryption and decryption algorithm to encrypt the latest difference parameter set △W';

[0191] Step G4, and sending the current institution parameter set ciphertext to the center server 11 through the corresponding first channel.

[0192] It should be noted that the division of the various modules of the above system is only a logical functional division, and in actual implementation, all or part of them can be integrated into one physical entity, or can be physically separated. These modules can all be implemented in the form of software called by a processing element; they can all be implemented in the form of hardware; or they can be partially implemented in the form of software called by a processing element and partially implemented in the form of hardware. For example, the central server can be a separately established processing element, or a chip integrated in a device, equipment or server. In addition, it can also be stored in the form of program code in the memory of the device, equipment or server and called and implemented by a processing element of the device, equipment or server to realize the corresponding processing function. The implementation of other modules is similar. In addition, all or part of these modules can be integrated together or independently implemented. The processing element described herein can be an integrated circuit with signal processing capability. In the implementation process, the processing steps of each module of the system can be completed by the integrated logic circuit of the hardware in the processing element or the instructions in the form of software.

[0193] For example, the modules of the above system can be one or more integrated circuits configured to implement the aforementioned processing steps, such as one or more application specific integrated circuits (ASICs), or one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs), etc. For another example, when a module of the above system is implemented in the form of program code called by a processing element, the processing element can be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules can be integrated together to be implemented in the form of a system on a chip (SOC).

[0194] In the foregoing embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the foregoing method embodiments are generated. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, Bluetooth, microwave, etc.) manner. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)) and the like.

[0195] The embodiment of the present application provides a kind of medical model federated learning framework data processing system, the system includes: center network, multiple one type of organization network, multiple two type of organization network;Center network is set in data center, one type of organization network is set in high-grade medical institution, two type of organization network is set in low-grade hospital institution;Center network and the federated learning framework of all one type of organization network belongs to vertical federated learning framework, one type of organization server under vertical framework has no interconnection relationship with each other, only to the center server connection;Center network and the federated learning framework of all two type of organization network belongs to horizontal federated learning framework, two type of organization server under horizontal framework can be interconnected with the service of other one type or two type of organization.The center server of center network and each organization server are carried out data transmission through independent secure data transmission channel, center server is used to periodically synchronize the model feature set of medical model and the public parameter set required by OT protocol to all organization servers, also be used to periodically distribute the latest version of model parameter of base model to all organization servers to start a round of model training task, also be used to aggregate, decrypt the organization parameter set ciphertext fed back by all organization servers in homomorphic decryption mode to obtain the latest version of model parameter of base model.The server of one type of organization is trained in each round of model training task Local model according to the sample set of the organization, and the organization model parameter obtained by training is encrypted based on homomorphic encryption mode, and the organization parameter set ciphertext obtained is sent to the center server.The server of two type of organization is first based on the PSI method realized from OT protocol in each round of model training task Supplemental features are obtained from other organization servers connected horizontally to improve the sample set of the organization, then the local model is trained based on the improved sample set, then the organization model parameter obtained by training is encrypted based on homomorphic encryption mode, and the organization parameter set ciphertext obtained is sent to the center server.The system of the embodiment of the present application expands the node access range and expands the training sample range on the one hand through the added horizontal federated learning framework;On the other hand, the transmission security of organization model parameter is improved through homomorphic encryption mechanism;Thirdly, through the PSI method realized from OT protocol, the transmission security of shared information of horizontal organization is guaranteed, and the privacy protection intensity of non-shared information is improved.

[0196] Those skilled in the art should further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware, computer software or a combination of both, and in order to clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in the above description. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0197] The steps of a method or algorithm described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module can reside in random access memory (RAM), flash memory, read-only memory (ROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such that the processor can read information from, and write information to, the storage medium. In the alternative, hard disk can be used as a storage medium.

[0198] The above detailed description has been described for clarity of understanding only. It should be apparent to those skilled in the art that many modifications, alternative constructions, improvements, and / or equivalents can be suggested by those skilled in the art and / or can be essential for implementing the present application depending upon specific application, the obvious modifications, changes, improvements, and / or equivalents being within the spirit and scope of the application.

Claims

1. A data processing system of a medical model federated learning framework, characterized in that, The system comprises a center network, a plurality of first-institution networks, and a plurality of second-institution networks; The center network corresponds to a data center; the center network comprises a center server and a base model; the center server is connected with the base model; the center server is also connected with first and second institution servers of each first and second-institution network respectively; The base model is a preset medical model; The center server is configured to establish a secure data transmission channel with each of the first and second institution servers according to a preset key agreement protocol, which is referred to as a corresponding first channel; The center server is also configured to periodically synchronize a model feature set of the medical model and a public parameter set required by an oblivious transfer protocol to all institution servers through all the first channels; The center server is further configured to periodically send the latest version of model parameters of the base model as a corresponding previous version of base parameters set; and generate a pair of public and private keys for homomorphic encryption and decryption calculation based on a preset parameter aggregation mode for the previous version of base parameters set, denoted as a corresponding public and private key 、 ; and distribute a model training notification carrying the public key and the previous version of base parameters set to all institution servers through all the first channels; the parameter aggregation mode includes a non-weighted mode and a weighted mode. The center server is further configured to, after receiving all the institution parameter set ciphertexts returned by all the institution servers through all the first channels, perform new version parameter aggregation processing based on the parameter aggregation mode, the private key , the previous version base parameter set, and all the institution parameter set ciphertexts to obtain a corresponding new version base parameter set; and save the new version base parameter set as the latest version model parameter of the base model. The first-institution network corresponds to a medical structure; the first-institution network comprises a first institution server, a first sample set, and a first local model; the first institution server is connected with the first sample set and the first local model respectively; The first local model has the same model structure as the base model but different model parameters; The first institution server is configured to store the received model feature set, public parameter set, and model training notification; The first institution server is also configured to perform a round of local training on the first local model based on the model training notification and the first sample set every time a new model training notification is received, and send the corresponding institution parameter set ciphertext to the center server; The second-institution network corresponds to a medical structure; the second-institution network comprises the second institution server, a second sample set, and a second local model; the second institution server is connected with the second sample set and the second local model respectively; each second institution server is also interconnected with another one or more first and second institution servers; The second local model has the same model structure as the base model but different model parameters; The second institution server is configured to store the received model feature set, public parameter set, and model training notification; The second institution server is further configured to extract the corresponding public key from the current model training notification upon receiving a new model training notification and the previous plate base parameter set; and parameter initialize the second local model based on the previous plate base parameter set; and perform missing feature checking based on the model feature set and the second sample set to obtain a corresponding missing feature set X; and perform sample feature completion processing on the second sample set through a round of interaction with other institution servers connected to itself according to the public parameter set and the missing feature set X to obtain a corresponding third sample set; and perform a round of local training on the second local model based on the third sample set to obtain a corresponding institution parameter set ciphertext and send the institution parameter set ciphertext to the center server.

2. The data processing system of the medical model federated learning framework according to claim 1, wherein The model feature set comprises a plurality of model input / output features, each of which corresponds to a type of medical feature information; The public parameter set of the inattentive transmission protocol includes cyclic group parameters G of a discrete logarithm problem, generator parameters g of a cyclic group, order parameters q of a cyclic group, a hash function H for mapping a patient identity to a cyclic group element ID , a hash function H for key derivation KEY ; the function input of the hash function H ID is a social identity of a patient, and the output is a group element in the cyclic group parameters G; the input of the hash function H KEY is a power operation result of a cyclic group element, and the output is a derived key of a specified key length; the social identity at least includes an ID card identity; The first sample set comprises a plurality of first sample records; each first sample record is a historical medical record of a patient, including the social identity of the current patient and all medical feature information of the current patient in the current medical structure; the corresponding patients of all first sample records are not repeated; the types of medical feature information at least include medical record text, physical examination report text, medical examination report text, medical examination report text, medical examination image, medical examination audio, treatment plan text, and diagnosis record text; All medical feature information of the first sample record covers all model input / output features of the model feature set; The second sample set comprises a plurality of second sample records; each of the second sample records is a historical medical record of a patient, and comprises the social identity of the patient and all the medical feature information of the patient in the current medical structure; The corresponding patients of all the second sample records are not repeated; All the medical feature information of the second sample records covers all the model input / output features of the model feature set; the types of the medical feature information of all the second sample records in the second sample set are fully aligned, and there is no difference in the type range of the medical feature information between any two second sample records; The third sample set comprises a plurality of third sample records; each of the third sample records is a medical record of a patient, and comprises the social identity of the patient and all the medical feature information of the patient in one or more medical structures; the corresponding patients of all the third sample records are not repeated; All the medical feature information of the third sample records covers all the model input / output features of the model feature set; said set of missing features X is not empty, by one or more missing features x j compositions, 1 ≤ index j ≤ N X , N X is the total number of missing features of the current set of missing features; each of said missing features x j corresponds to one of said pairs of model input / output features in said set of model features.

3. The data processing system of the medical model federated learning framework according to claim 1, wherein the key agreement protocol comprises at least a DH protocol, an ECDH protocol, an IKEv2 protocol, and a TLS-PSK protocol; The two servers on both sides of each first channel have a pair of shared keys; The transmission mechanism of each first channel is that the center server or the first and second institution servers as the current channel data sender encrypts the original plaintext based on the shared key and sends the obtained ciphertext to the data receiver on the other side of the current channel, and the first and second institution servers or the center server as the current channel data receiver decrypts the ciphertext based on the shared key to obtain the corresponding original plaintext.

4. The data processing system of the medical model federated learning framework according to claim 1, wherein the parameter aggregation mode is identified.

5. The data processing system of the medical model federated learning framework according to claim 1, wherein the parameter aggregation mode is identified. The center server is specifically configured to generate a pair of public and private keys for homomorphic encryption and decryption calculation for the pre-version base parameter set based on the preset parameter aggregation mode, denoted as a corresponding public and private key , .

6. The data processing system of the medical model federated learning framework according to claim 1, wherein the first institution server is specifically configured to, when the first institution server sends the corresponding institution parameter set ciphertext to the center server after performing one round of local training on the first local model based on the model training notification and the first sample set: If the parameter aggregation mode is a non-weighted mode, a preset semi-homomorphic algorithm is used to generate a public key and a private key corresponding to the public key ; the semi-homomorphic algorithm at least includes a Paillier algorithm; If the parameter aggregation mode is a weighted mode, a preset fully homomorphic algorithm is used to generate a public key and a private key corresponding to the public key and the private key respectively; and the fully homomorphic algorithm at least supports BFV algorithm, BGV algorithm, CKKS algorithm, FHEW algorithm and TFHE algorithm. and performs parameter initialization on the first local model based on the previous version base parameter set; The center server is specifically configured to, when performing new edition parameter aggregation processing on the basis of the parameter aggregation mode, the private key , the previous edition base parameter set, and all the institution parameter set ciphertexts to obtain a corresponding new edition base parameter set, The total number of the institution parameter set ciphertexts is counted to obtain a corresponding total number N; and each of the institution parameter set ciphertexts is recorded as a corresponding ciphertext EW i , 1≤index i≤N; and the former edition base parameter set is recorded as a corresponding parameter set W old ; and the new edition base parameter set is recorded as a corresponding parameter set W new ; and performs one round of training on the first local model after initialization based on the first sample set; If the parameter aggregation mode is a non-weighted mode, the corresponding new version of the differential parameter set ciphertext EW is generated in the manner of new ​ If the parameter aggregation mode is a weighted mode, the corresponding new version of the differential parameter set ciphertext EW is generated in a manner of new α i is a preset weight coefficient corresponding to the i-th ciphertext EW i . And based on the private key For the new version of the differential parameter set ciphertext EW new Homomorphic decryption yields the corresponding new version of the difference parameter set △W new ; , A preset homomorphic decryption function is used to decrypt the private key. For the new version of the differential parameter set ciphertext EW new Decrypt; and based on said parameter set W old and said new differential parameter set AW new calculating a corresponding said parameter set W new , . and sends the current institution parameter set ciphertext to the center server through the corresponding first channel.

7. The data processing system of the medical model federated learning framework according to claim 2, wherein extracting a corresponding public key from the model training notification and the previous set of parameters of the plate base ​ ​ and record the current model parameter set of the first local model as parameter set W at the end of the current round of training now , record the previous version base parameter set as parameter set W old , and calculate a latest difference parameter set AW = W now , W old based on the parameter set W now old ;​ and based on the public key encrypting the difference parameter set AW to obtain the corresponding institution parameter set ciphertext; Institutional parameter set ciphertext = , A preset homomorphic encryption function, used to employ the public key. The latest differential parameter set △W is encrypted; ​ ​ The second institution server is specifically configured to identify all medical features covered by the second sample set to obtain a corresponding medical feature set when the missing feature set X is obtained based on the model feature set and the second sample set; and identify the feature intersection of the medical feature set and the model feature set to obtain a corresponding current feature intersection; and the feature set in the model feature set except the current feature intersection is taken as the corresponding missing feature set X.

8. The data processing system of the medical model federated learning framework according to claim 2, wherein The second institution server is specifically configured to perform a round of interaction with other institution servers connected with itself according to the public parameter set and the missing feature set X, and perform sample feature supplement processing on the second sample set to obtain a corresponding third sample set through the round of interaction; performing a round of iteration on all the second sample records of the second sample set; and during the current round of iteration, taking the second sample record being currently iterated as a corresponding current sample record; and in the current sample record, adding N X j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x j corresponding to the missing feature x <000002 and perform a round of traversal on all the first and second institution servers connected with itself; and in the round of traversal, the first or second institution server currently traversed is taken as a current sender server; and based on the private set intersection method implemented by the inadvertent transmission protocol, sample missing features are obtained from the sender server according to the public parameter set and the missing feature set X to fill the third sample set; and at the end of the round of traversal, the latest third sample set is taken as the processing result of the sample feature supplement processing.

9. The data processing system of the medical model federated learning framework according to claim 8, wherein The second institution server is specifically configured to send an interaction request carrying the missing feature set X to the sender server when filling the third sample set based on the private set intersection method implemented by the inadvertent transmission protocol according to the public parameter set and the missing feature set X.

10. The data processing system of the medical model federated learning framework according to claim 9, wherein counting the total number N of the third sample set U ; and taking each of the social identity identifiers of the third sample set into the hash function H ID ; and calculating and taking the calculation result as the corresponding identification element h u = H ID (ID u ), 1≤index u≤N U , ID u is the u-th social identity identifier of the third sample set, h u ∈G; and set a corresponding remainder ring parameter Z based on the order parameter q q {0,1,2,…,q-1} and the remainder ring parameter Z q is performed N U times, and each selected random number is taken as a corresponding patient private key S u , S u ∈Z q ; and a corresponding patient public key C u is generated based on the generating element parameter g and each patient private key S u and the corresponding identification element h u , , C u ∈G. and after receiving the sender public key H sent back by the sender server, sending a patient public key set {C U} composed of N u patient public keys C u to the sender server; and receiving a ciphertext matrix {ED u,v} sent back by the sender server; the ciphertext matrix {ED u,v} is composed of N U × N V ciphertext data ED u,v ; N V is the total number of records of the first or second sample set of the sender server; and based on the hash function H KEY , the sender public key H and each of the patient private keys S u generate a corresponding patient shared key K u , ; and based on each of the patient shared keys K u decrypts the corresponding N V ciphertext data ED u,v obtains the corresponding N V plaintext data DED u,v , , f DE is a decryption function of a preset symmetric encryption and decryption algorithm; And the N U ×N V The plaintext data DED u,v is traversed for one round; and in the process of the current round, the plaintext data DED u,v is traversed as the corresponding current plaintext; and assembling rules of the preset patient missing information set and the check code, extract the corresponding patient missing information set D from the current plaintext u,v and the check code MAC u,v ; and based on a preset check code algorithm, calculate the check code of the patient missing information set D u,v to obtain the corresponding current check code MAC now ; and when the current check code MAC now matches the MAC u,v , the current patient missing information set D u,v is taken as the patient missing information set D u corresponding to the current plaintext, and based on the current patient missing information set D u , set N X medical feature information corresponding to the missing feature set X in the u-th third sample record of the third sample set. After receiving one of the interaction requests, the sender server takes the second institution server sending the current interaction request as a current receiver server; and extracts the corresponding missing feature set X from the current interaction request; and takes the local first or second sample set as a corresponding current sample set; and sends the sender public key H back to the receiver server. and the total number N of the corresponding total number of records of the current sample set is counted V ; and bringing each of the social identity identifiers of the current sample set into the hash function H ID performing the calculation and the result of the calculation as the corresponding identification element h v = H ID (ID v ), 1 ≤ index v ≤ N V , ID v is the vth social identity identifier of the current sample set, h v ∈ G; and set a corresponding remainder ring parameter Z based on the order parameter q q {0,1,2,…,q-1} and the remainder ring parameter Z q random number selection and the selected random number as a corresponding sender private key S, S∈Z q ; and generating a corresponding sender public key H based on the generator parameter g for the sender private key S, , H∈G; 11. The data processing system of the medical model federated learning framework according to claim 1, wherein And after sending the sender public key H, receiving the patient public key set {C u} sent by the receiver server; the patient public key set {C u} is composed of N U patient public keys C u ​ and based on said hash function H KEY , said set of patient public keys {C u}, said sender private key S and all said identification elements h v generate a corresponding patient shared key K u,v , ; and the first or second sample record corresponding to the vth social identity label in the current sample set is taken as the missing feature information d corresponding to the missing feature x j corresponding to the medical feature information v,j , if the sample record of the current sample set also does not have a certain missing feature x j corresponding to the medical feature information is taken as the missing feature information d v,j corresponding to the medical feature information is taken as the missing feature information d X corresponding to the medical feature information is taken as the missing feature information d v,j corresponding to the medical feature information is taken as the missing feature information d v {d v,j} And based on the preset checksum algorithm, the missing information set D of each patient is... v {d v,j } Perform checksum calculation to obtain the corresponding checksum MAC. v ; and from each of the sets of missing information D of the patients v {d v,j} and the corresponding check code MAC v composing a corresponding patient data DM v ; and based on each said patient shared key K u,v encrypting the corresponding said patient data DM v obtaining the corresponding said encrypted data ED u,v , , f EN is a preset encryption function of a symmetric encryption and decryption algorithm, f EN and f DE are inverse operation functions of each other. and the obtained N U ×N V ciphertext data ED u,v composing a corresponding ciphertext matrix {ED u,v} to the receiving server. The second institution server is specifically configured to send the corresponding institution parameter set ciphertext back to the center server when performing a round of local training on the second local model based on the third sample set to obtain the corresponding institution parameter set ciphertext; perform a round of training on the second local model which has completed initialization based on the third sample set; and send the current institution parameter set ciphertext to the center server through the corresponding first channel. and record the current model parameter set of the second local model as parameter set at the end of the current round of training , record the previous version of the base parameter set as parameter set , calculate the latest difference parameter set △W ’ = - ; and based on the public key to the difference parameter set AW ’ and the corresponding institution parameter set ciphertext is obtained by encryption. Institution parameter set ciphertext , The preset homomorphic encryption function is used to encrypt the latest difference parameter set △W using the public key ’ . ​

Images