Pixel-based Homomorphic Residue Number System Scheme for Privacy-Preserving Neuroimaging Datasets Encryption and Decryption

Abstract

Recently, there has been some interest in the development of a homomorphic privacy-preserving classification method for neuroimages based on the residue number system (RNS) and deep CNN. This paper describes the RNS homomorphic encryption system for neuroimages and analyses its security efficiency in relation to the moduli set. The proposed system's security efficiency is evaluated using the histogram, key space, key sensitivity, and correlation analysis. The analysis results show that the proposed RNS homomorphic scheme is a fully homomorphic encryption (FHE) scheme capable of encrypting and decrypting neuroimages without sacrificing any inherent neural-biomarker features. The results also show that the scheme is resistant to statistical attacks like histogram, brute-force, correlation coefficient, and key sensitivity. Therefore, the proposed RNS-FHE scheme can be applied to any type of neuroimaging dataset and is suitable for the design of homomorphic privacy-preserving methods compared to the best-known state-of-the-art.