A privacy-preserving computer-aided diagnostic method based on inner product function encryption

Abstract

This invention discloses a privacy-preserving computer-aided diagnostic method based on inner product function encryption. The steps include: 1) A key generation center calls a group generation algorithm to generate a master key msk and public parameters pp; 2) Calculates and returns a key to the data owner based on msk, the data owner's identifier ID, and the input vector; 3) The model owner calculates D = e(h1, h2) based on pp, ID, and the input vector. r And then send the ciphertext to the computer-aided diagnostic device; 4) The data owner obtains the inner product operation result based on the ciphertext and the key, and then calculates the prediction result.

Description

Technical Field

[0001] This invention belongs to the field of bioinformatics, specifically a privacy-preserving computer-aided diagnostic method based on inner product function encryption. Background Technology

[0002] The rapid development of computer-aided diagnosis based on machine learning models has profoundly impacted modern medical activities. Computer-aided diagnosis can help doctors make accurate diagnoses, thereby providing better medical services to patients. Simultaneously, its large-scale adoption can significantly improve healthcare in remote and underdeveloped areas, promoting public health. However, current computer-aided diagnostic systems face a series of challenges regarding data privacy and model ownership. For example, in these systems, owners of medical models must disclose model parameters or access user data used for prediction tasks in medical scenarios. However, user data often contains sensitive and private information; furthermore, machine learning models trained by enterprises or hospitals are important assets, and the leakage of model parameters can cause losses to these enterprises, thus they should not be arbitrarily accessed by third parties. Therefore, ensuring both user privacy and model parameter security during computer-aided diagnosis, allowing model owners, diagnostic system providers, hospitals, and users to conduct medical activities while safeguarding their respective interests, is a crucial prerequisite for promoting the large-scale application of computer-aided diagnosis.

[0003] Privacy-preserving model prediction

[0004] To ensure data privacy for participants during machine learning model prediction, there are currently three main technical approaches: Differential Privacy (DP), Homomorphic Encryption (HE), and Secure Multi-party Computation (MPC). Differential Privacy (DP) was first proposed by Dwork in 2006 (Dwork, C. "Differential Privacy. Automata, Languages ​​and Programming. 2006." 1-12.). Addressing the privacy-preserving prediction problem, Dwork et al. proposed the concept of privacy-preserving prediction based on DP (Dwork, Cynthia, and Vitaly Feldman. "Privacy-preserving prediction." Conference On Learning Theory. PMLR, 2018.). This involves introducing random numbers from a specific distribution to add noise to the original model parameters, thereby enabling the calculation of prediction results without knowing the original model parameters. However, the random noise introduced in DP-based schemes often significantly impacts the model's prediction accuracy, making such schemes difficult to apply in real-world scenarios. In HE-based privacy-preserving prediction, the user of the model prediction results first generates a key pair (pk1, sk1) for HE and sends the public key pk1 to the model owner. Then, the user encrypts the original data using pk1 and sends the resulting ciphertext... Send to the model owner. Upon receiving the ciphertext, the model owner calculates... The model owner generates prediction results and sends the encrypted results to the data owner. Finally, the data owner decrypts the encrypted data to obtain the prediction results. However, this process involves a series of encryption and decryption operations, as well as homomorphic multiplication and addition operations, which are very time-consuming. In MPC-based prediction, the model owner and data owner share secret fragments of their local data through an arithmetic secret sharing protocol. That is, the model owner calculates... in It is a vector of model parameters. It is a random vector. The data owner calculates it in the same way. Then, the data owner and the model owner exchange shares of their local data (the model owner receives the data). Data owner obtains They jointly compute by performing MPC multiplication and summation protocols sequentially. and The inner product of the data is then used. Finally, the data owner reconstructs the prediction result using the intermediate results. In MPC-based prediction, all intermediate results are calculated on random numbers, so there is no information leakage about the original data. However, although the computational overhead is relatively small, the communication cost is huge. Therefore, in both HE-based and MPC-based prediction, all participants need to be online in real time to complete the prediction process, and the computational and communication costs of this process are very high. Summary of the Invention

[0005] To address the problems existing in the prior art, the main objective of this invention is to propose a privacy-preserving computer-aided diagnostic method based on inner product function encryption. This method is used for the outsourcing and privatization of machine learning models in computer-aided medical diagnostic scenarios, while ensuring the security of model parameters. The method proposed in this invention focuses on two main characteristics: First, users can obtain prediction results without interacting with the model owner. Second, the owner of personal data can flexibly authorize relevant third parties to access the prediction results. For example, in a company that produces computer-aided diagnostic medical models, when a patient seeks treatment in a hospital, the attending physician needs to access the model and obtain prediction results using the patient's data (such as examination results and electronic medical records). If further diagnosis is needed, the doctor needs to authorize the current prediction results as a reference for other relevant doctors. In this scenario, the company is the model owner, the first doctor is the user of the prediction results, and subsequent doctors are users who need authorization to access the previous prediction results.

[0006] The technical solution of this invention is as follows:

[0007] The privacy-preserving computer-aided diagnostic method based on inner product function encryption includes the following steps:

[0008] 1) System initialization: The key generation center calls the group generation algorithm to generate... in Let be a multiplicative cyclic group, e be a bilinear mapping, and p be the order of the multiplicative cyclic group. Select random numbers. Calculate g T =e(g1,g2). Select hash function h: ( (representing the group of non-zero integers of order p) and random numbers ( (where p is an integer group), calculate f(·) = vh(·). The master key can be represented as: msk = (s, t, h), and the common parameters can be represented as: pp = (e, g1, g2, g T ,h1,h2,k1,f).

[0009] 2) Key generation and distribution: Based on the master key msk from step 1), the data owner's identification information ID, and the input vector... (in This represents the feature vector corresponding to a user's personal information, where each element x... i The key generation center calculates the feature value of the i-th feature (where n is the total number of features). And return the key Assign the data to the owner corresponding to the ID. For authorized users, assuming their identification information is ID′, the key generation center calculates... And send Give it to the authorized user.

[0010] 3) Model parameter encryption and model outsourcing deployment: Based on the public parameter pp from step 1), the data owner's identification information ID, and the input vector... (in This represents the parameter vector owned by the model owner, where each element θi represents the input vector. The weights corresponding to the i-th feature in the model are selected by the model owner using random numbers. And calculate D = e(h1, h2) r For i∈{1,2,…n}, calculate And send the encrypted data corresponding to the model parameters. To the computer-aided diagnostic device designated by the data owner, perform step 4); when the data owner designates an authorized user, the encrypted data will be sent to the device. Send to the proxy server specified by the data owner and proceed to step 5).

[0011] 4) Privacy-preserving model prediction. Based on the model parameters in step 3), the encrypted data is... and the key in step 2). The data owner first obtains the result of the inner product operation. Then calculate This is the prediction result.

[0012] 5) Authorization of prediction results by proxy.

[0013] a) Assuming the authorized user's identification information is ID′, based on the data owner's private key in step 2), The data owner first selects a random number. And calculate Then select a random number. And calculate make The re-encryption key can be represented as rk ID→ID′ = (d′1, d′2, d′3). The final data owner sends rk. ID→ID′ The proxy server designated by the data owner.

[0014] b) The proxy server utilizes rk from step a). ID→ID′ For the encrypted model parameters in step 3), Perform re-encryption: calculation Let D′=d′3, then the re-encrypted ciphertext can be represented as send Give it to the authorized user.

[0015] c) The authorized user obtains the information in step b). And using the key from step 2). Perform decryption operation: Calculate Then calculate

[0016] d) Calculation by authorized users This is the prediction result.

[0017] The beneficial effects of this invention are:

[0018] Based on the aforementioned privacy-preserving computer-aided diagnostic framework, this invention achieves privacy-preserving model prediction and proxy authorization of prediction results, while also realizing encrypted access control based on identification information (ID).

[0019] To address the privacy concerns of model parameters and user data in computer-aided diagnosis, this invention proposes a novel privacy-preserving prediction framework applicable to computer-aided diagnosis in scenarios where models are outsourced. The proposed framework utilizes an identity-based inner product proxy re-encryption scheme. This scheme introduces a flexible proxy authorization feature and incorporates the concept of access control based on identifier information, allowing for authorization verification through user identifier information while ensuring the security of both the original data and the prediction results. Attached Figure Description

[0020] Figure 1 This is a flowchart of the key distribution process.

[0021] Figure 2 Deployment flowchart for model encryption.

[0022] Figure 3 A flowchart for privacy protection prediction.

[0023] Figure 4 Flowchart for the result proxy. Detailed Implementation

[0024] This invention realizes privacy-preserving model prediction and prediction result proxy authorization, and the specific implementation method is as follows.

[0025] System initialization and key distribution

[0026] The system initialization and key distribution process is as follows: Figure 1 As shown, the key generation center first calls the system initialization algorithm, inputs system security parameters, and generates the necessary public parameters for system operation, as well as the master key used to generate user private keys. Subsequently, the key generation center receives the user's identifier and vector information. If the received vector is empty, the user is the predictor of the result; otherwise, the user is the data owner. The key generation center generates the corresponding private key based on the different user types and distributes it to each user.

[0027] Model Encryption Deployment

[0028] Model encryption deployment process as follows Figure 2 As shown, the model owner first trains the machine learning model locally to obtain the corresponding model parameters, then encrypts the model parameters to obtain the corresponding ciphertext data, and finally sends the ciphertext parameters to the cloud server or the model user's computer for deployment.

[0029] Privacy Protection Prediction

[0030] Privacy protection prediction process as follows Figure 3 As shown, the user who needs to make a prediction first obtains the encrypted model parameters, then calls the decryption algorithm to decrypt the encrypted text and obtain the inner product operation result. Finally, the user further calculates the model prediction result based on the inner product result.

[0031] Result of Agency Authorization

[0032] Result of proxy authorization, such as Figure 4 As shown, the authorization process mainly involves three parties: the data owner, the cloud server, and the authorized user. It is assumed that the encrypted model parameters have been deployed to the cloud server, and both the data owner and the authorized user have obtained their private keys. The data owner first calculates a re-encryption key using their private key and the authorized user's ID information, and sends this key to the cloud server. The cloud server performs a re-encryption operation using the received re-encryption key and the encrypted model parameters, obtaining the re-encrypted model prediction result, and sends this result to the authorized user. Upon receiving the encrypted prediction result, the authorized user decrypts it using their private key to obtain the corresponding inner product operation information. Finally, the authorized user calculates and obtains the model prediction result locally based on the inner product result.

[0033] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Those skilled in the art can modify or make equivalent substitutions to the technical solutions of the present invention without departing from the spirit and scope of the present invention. The scope of protection of the present invention should be determined by the claims.

Claims

1. A privacy-preserving computer-aided diagnostic method based on inner product function encryption, comprising the following steps: 1) Key Generation Center calls group generation algorithm to generate ; wherein, is a multiplicative cyclic group, is a bilinear map, is the order of the multiplicative cyclic group; select a random number , calculate ; select a hash function and a random number ; calculate , get the master key and public parameters ; 2) a key generation center calculates a key based on the master key and the identity information of the data owner ​​​​​​​​​ 3) The model owner selects a random number. Then, based on the common parameters Identification information of the data owner and input vector calculate as well as , , For input vectors The i-th element represents the input vector. The Middle The weights corresponding to each feature, where n is the total number of features; then the ciphertext corresponding to the model parameters... Send to the computer-aided diagnostic device designated by the data owner; 4) The data owner, based on the encrypted text and keys Obtain the result of the inner product operation. Then calculate , This is the prediction result.

2. The method according to claim 1, characterized in that, In step 2), when the data owner designates an authorized user, the identifier information of the authorized user is: Key generation center calculation and send The method by which the authorized user obtains the prediction result is as follows: a) The data owner selects a random number. And calculate Then select a random number. And calculate ,make Obtain the re-encryption key And send it to the proxy server specified by the data owner; b) The proxy server equipment uses the received... For the ciphertext Perform re-encryption: calculation and order The ciphertext obtained by re-encryption And send it to the authorized user; c) The authorized user, based on the received... and key Perform decryption operation: Calculate Then calculate d) The authorized user's calculation , This is the prediction result.

3. The method according to claim 1, characterized in that, The input vector The parameter vector and input vector belong to the model owner. The i-th element For the first The weights corresponding to each feature.

4. The method according to claim 1 or 2, characterized in that, The hash function random numbers random numbers , random numbers ; for non-zero integer group of order, for A group of integers of order 1.

Images