Research On Methods Of Privacy-Preserving For Medical Data Query Computation

With the continuous and rapid development of network technology and wearable devices,medical data has shown explosive growth.As an important asset,medical data,by using the mathematics,statistics and computational science and some other technologies to convert data into medical knowledge,which helps the users and medical workers to make accurate and efficient decision support.However,security and privacy issues are becoming the block for using medical data to make an intelligent decisions.The European Union's General Data Protection Regulation(GDPR)is a symbol,when the major countries in the world have strengthened legislation on data security and privacy protection,the corresponding technical protection methods have also become research hotspots.Privacy computing refers to a computing theory and methodology,which includes all computing operataions of data generation,storage,processing,publishing,and destruction during the entire lifecycle of privacy data.Since medical data usually needs to be processed and applied to generate value,while applications need to provide users with query computation services,which is easy to cause sensitive data leakage.As an important research topic,the privacy-preserving methods for medical data query computation will be the core of this dissertation.In a typical medical data query computation system,medical service providers usually use data mining algorithms to generate medical knowledge models from medical data,and provide users with online medical data query computation services.For the sake of intellectual property,medical service providers have to protect the unique medical knowledge from leaking during the process of medical data query computation;at the same time,for some privacy-preserving reasons,users also hope that their medical data will not be stolen by the medical service providers during the medical data query computation.In this dissertation,we will focus on security and privacy issues during the process of medical data query computation.Considering the several characteristics of medical data,such as sensitivity,volume,value and dynamic,we first analyzed the query computation system's application requirements,including efficiency,collaboration,and real-time service,then conducted extensive research on the existing medical data privacy-preserving mechanism.After that,we selected three typical different interaction scenarios of medical data query computation,and applied the theories and methods of cryptography,machine learning,and mathematical statistics to construct three different secure and efficient privacy-preserving schemes.At last,we verified the safety and efficiency of the proposed privacy-preserving schemes by using theoretical analysis and prototype system testing.Specifically,the main contribution of this dissertation includes the following three aspects.1)Privacy-preserving method for medical pre-diagnosis in two-party interactionConsidering the huge scale and high-dimensional characteristics of medical data,we constructed a novel medical knowledge model by using Skyline Query,and verified the high accuracy of the model with a real medical dataset.On this basis,to solve the privacy data leakage issues during the medical data query computation under the two-party interaction scenario,we converted the integers comparison problem into the set intersection problem by using 0-1 Encoding Technology at first;then encrypts and permutes the user's medical query data and medical knowledge model with Polynomial Aggregation and Multi-party Random Permutation technologies.In this way,we construct a fast and secure two-party query computation menchanism,and then design an efficient medical pre-diagnosis privacy-preserving scheme,which can satisfy the security and privacy requirements of medical knowledge model and user medical query.Through detailed security analysis and extensive testing in actual application scenarios with real data sets,the results show that the scheme can achieve security and efficiency.2)Privacy-preserving method for medical pre-diagnosis in three-party interactionTo improve the accuracy of medical diagnosis models,we consider a Medical Alliance,which collaborates with several medical data centers to generate a medical diagnosis model,by using the additivity of Skyline Query.Aiming at the privacy data leakage problem during the medical diagnosis service in the cloud platform,we proposed an efficient and privacy-preserving medical diagnosis scheme in a three-party interaction scenario.Specifically,by using the Chinese Remainder Theorem,we decompose the decrypted private key into two parts,and distribute them to the cloud computing platform and Medical Alliance;then calculate the medical query data and medical knowledge model in ciphertext form with Paillier homomorphic encryption;at last,Medical Alliance obtains the medical pre-diagnosis results by using the partial decryption technology.The proposed scheme achieves the confidentiality of the medical knowledge model of the Medical Alliance and the privacy of user medical query data.Finally,we prove the safety with analysis,and develops a prototype system to verify the accuracy and efficiency of the proposed scheme by testing with real medical data sets.3)Privacy-preserving method for healthcare monitoring over outsourced cloudTo improve the misjudgment of healthcare monitoring caused by the dynamic characteristic of medical data,we construct a novel healthcare monitoring model by using the Decision Tree Classifier,which containing multiple physiological characteristics and corresponding normal value intervals.Compared with the traditional decision tree monitoring models,our medical monitoring model has higher accuracy.On this basis,to protect the privacy data during the process of medical data query computation over the outsourced cloud,the medical service provider needs to encrypts the interval parameters of the decision tree monitoring model and outsource them into the cloud platform at first,then generates some intermediate results which was computed by all subsets of a large integer interval with the decryption keys,the intermediate results were stored with the Bloom filter.After that,the cloud platform performs the query computation between the user's medical query and the medical knowledge model under the ciphertext format.At last,the user obtains the specific intermediate results by using the private decryption key,and then matches the intermediate results in the Bloom filter.With these operations,we constructed a fast range query computation mechanism under ciphertext,and then designed an efficient healthcare monitoring privacy-preserving scheme.We also verified the accuracy of the proposed scheme with the real diabetes data set,and tested the performance in the actual environment.The results showed that our scheme can achieve high accuracy and efficiency.