Privacy-Preserving Pre-diagnosis over Single-Label Medical Records

This chapter proposes an accurate and privacy-preserving pre-diagnosis scheme over outsourced single-label medical records. When the proposed scheme is applied, medical institutions can securely outsource pre-diagnostic models to a cloud server, allowing them to provide timely and convenient services to patients while guaranteeing the confidentiality of sensitive data, including the parameters of models and queries from patients. Specifically, the pre-diagnostic model used in the proposed scheme is a k-nearest neighbor (kNN) classifier whose similarity metric is Mahalanobis distance (MD). Accordingly, an MD-based similarity secure comparison algorithm (MDSC) is developed by improving the noise-added matrix encryption method (NMEM). As the core module of the proposed scheme, MDSC enables the pre-diagnostic model to run on ciphertexts with linear complexity. To further decrease the computational overhead, an MD-based hierarchical index tree is introduced to the scheme. Finally, detailed security analysis and extensive experimental tests demonstrate the proposed scheme can withstand closeness-same-pattern chosen-plaintext attacks while offering high pre-diagnosis accuracy and query efficiency.