Research On Key Techniques Of Privacy-preserving Services Over Medical Data

The rapid development of the medical Internet of Things has continuously improved the level of informatization in the medical and health field,and resulted in an explosive growth of medical data.In order to mine the potential value of medical data,various online medical services based on different data analysis technologies have emerged.However,medical data is highly sensitive,it usually involves personal life and health information that should be kept secret.If medical data is directly sent to service providers who cannot be fully trusted for services,it will bring serious issues of data security and privacy leakage.Focusing on the above problem,a number of laws and regulations have been promulgated at home and abroad to help establish and improve the rules and regulations for the protection of users’ sensitive information.Meanwhile,academic researchers are also paying attention to privacy preservation in the process of medical services,based on homomorphic encryption,secret sharing,garbled circuits and other encryption techniques,they have proposed many privacypreserving schemes for various medical services.However,the existing privacy-preserving schemes either sacrifice huge resources for computation/communication to ensure high security,or make a considerable compromise in security to provide high efficient services,they cannot strike a good balance between security and performance.Besides,most of them choose to directly use general data mining methods to provide services to users,but the popular data mining methods may not be applicable to all types of medical data.Therefore,for different types of medical data,under the premise of protecting privacy,how to provide accurate and efficient medical service is still a challenging research topic.In this dissertation,by analyzing different data characteristics to improve the accuracy of service models,applying efficient encryption techniques to ensure the security of sensitive information,and introducing retrieval optimization structures to narrow the scope of queries,we construct a series of accurate,efficient and privacy-preserving online medical schemes over gene sequences,medical images,and electronic medical records.Moreover,the security,accuracy and efficiency of the proposed schemes have been verified from the perspective of theoretical analysis and experimental testing.Specifically,the main contributions of this dissertation include the following four aspects:(1)In view of efficient and privacy-preserving similar patients query services,we first design a data structure named genetic BK-tree based on our improved approximate edit distance computation method,which can achieve fast and accurate retrieval of similar gene sequences.Then,by combining genetic BK tree with hash message authentication code and symmetric key encryption algorithm,an efficient and privacy-preserving scheme is constructed over outsourced gene sequences.Finally,we prove that the proposed scheme can resist selectively chosen-plaintext-attack,and the extensive experimental results conducted on real and synthetic datasets demonstrate its accuracy and high-efficiency.(2)In view of accurate and privacy-preserving similar medical image retrieval services,we first design a Privacy-preserving Mahalanobis Distance(MD)Comparison method under one cluster based on an enhanced matrix encryption technology,named PMDC.Then,by introducing fuzzy C-means clustering algorithm and applying PMDC and improved PMDC which can securely compare MD under different clusters,an accurate and privacy-preserving scheme is constructed over outsourced images and indexes.Finally,we prove that the proposed scheme can resist known-plaintext-attack,and the extensive experimental results conducted on real and synthetic datasets demonstrate its high-accuracy and efficiency.(3)In view of accurate and efficient online medical pre-diagnosis services,we first design a Secure MD-based similarity Comparison method under the same covariance matrix by using a learning-with-errors-based matrix encryption algorithm,named SMDC.Then,by combining hierarchical index tree with SMDC and improved SMDC which can securely compare MD under different covariance matrices,an accurate and efficient scheme is constructed over outsourced encrypted electronic medical records.Finally,we prove that the proposed scheme can resist closeness-same-pattern chosen-plaintext-attack,and the extensive experimental results conducted on real and synthetic datasets demonstrate that it can achieve higher accuracy and efficiency than existing similar schemes.(4)In view of efficient and privacy-preserving multi-disease simultaneous diagnosis service,we first choose multi-label k nearest neighbor classifier as the pre-diagnosis model to train the stored multi-label electronic medical records.Then,by introducing k means clustering algorithm to improve query speed and utilizing an efficient secure two-party inner product calculation protocol to protect privacy,an efficient and privacy-preserving scheme is constructed under a two-party computation model.Finally,we prove that the proposed scheme has the ability to protect privacy,and the extensive experimental results conducted on real and synthetic datasets demonstrate that its diagnostic results are accurate and its performance is efficient.