Privacy-Preserving Pre-diagnosis over Multi-label Medical Records

This chapter introduces an efficient and privacy-preserving pre-diagnosis scheme over multi-label medical records. Given a medical record from a new patient, the scheme aims to assist physicians in quickly pre-diagnosing which diseases the patient may have while keeping medical records confidential. Specifically, we first design a novel privacy-preserving pre-diagnosis scheme named CREDO based on multi-label k-nearest neighbors (ML-kNN) classification algorithm and two-party inner product computation protocol. Then, we improve the pre-diagnosis efficiency of CREDO by adapting K-means clustering to reduce the search space of ML-kNN. With CREDO, physicians can ensure sensitive medical records are secure, and service providers can provide high-efficiency service without revealing pre-diagnosis models. Finally, rigorous security analysis demonstrates CREDO can guarantee security under different threats, and extensive performance evaluations show the scheme is highly accurate and efficient.