Ensure privacy and security in the process of medical image analysis

Abstract

With fast growth of cloud computing service, data security becomes a pressing need to ensure users' privacy. In this paper, we propose a homomorphic encryption method that enables direct operation over the encoded data, and hence facilitates the complete privacy protection. The method roots in residue number system and takes advantage of the parallelism in cloud computing. Experiments were conducted with magnetic resonance images for filtering-based image operations and arithmetic operations. It is evident that the encrypted images (shares) are nearly random in their contents and have almost no correlation with the original image. Most importantly, the same processing steps applied to the ciphertext result in almost identical image as the one generated without encryption following the same steps. Due to redundancy introduced in the encryption phase, the computational cost and data volume increase. However, by taking in account the parallel computing power in cloud service we expect much shorter time, which is acceptable in off-line applications.