移动医疗网络中隐私保护密文计算研究

Privacy-preserving is one of the main security requirements of mobile healthcare networks. The most common method to protect data privacy is to encrypt the data. However, computation on the encrypted data faces challenges. In order to solve this problem, homomorphic encryption with the capacity of both encryption and computation is studied. We avoid partial homomorphic encryption and full homomorphic encryption, and define the differentiated levels of circuit depths, and improve modulus switching to reduce the noise in ciphertext, so a differentiated somewhat homomorphic encryption scheme is proposed. We analyze the regression and classification models which are usually used to predict disease risks, and introduce modularize functions as intermediate state to form the expression of circuit-function-protocol, thus the computation protocol on encrypted data is designed based on the differentiated somewhat homomorphic encryption scheme. Constructing a prototype system of disease risk prediction on mobile healthcare networks, privacy-preserving disease risk prediction is achieved. The characteristics of privacy-preserving and computation on encrypted data are analyzed, and the feasibility of our proposed scheme and protocol is verified. This study of privacy-preserving computation on encrypted data will contribute to the secure development of mobile healthcare networks, and the research and application of homomorphic encryption. Therefore, this project is valuable for theory and practice.

移动医疗网络安全需求的核心是隐私保护，加密是保护医疗数据隐私的最常用方法，但数据在密文形态下的计算问题却是安全研究所面临的挑战。为了解决这个问题，本课题研究兼具加密与计算双重能力的同态加密，避开部分同态加密所引入的交互问题与全同态加密的效率问题，根据计算需求定义电路深度的差异化等级，改进模交换技术约减等级间的密文噪声，设计差异化Somewhat同态加密方案；分析疾病风险预测所用的回归与分类运算，引入模块化函数作为中间态，建立电路-函数-协议的运算表达，基于差异化Somewhat同态加密方案实现医疗数据密文计算协议；构建轻量级移动医疗网络及疾病风险预测系统，实现隐私保护的疾病风险预测计算，分析系统的隐私保护与密文计算特性，验证所构造方案与协议的可行性。本课题研究移动医疗网络的隐私保护密文计算，有助于推进移动医疗网络的安全发展，丰富同态加密技术的研究与应用，具有重要的理论意义与实践价值。

移动医疗网络安全需求的核心是隐私保护，加密是保护医疗数据隐私的最常用方法，但数据在密文形态下的计算问题却是安全研究所面临的挑战。为了解决这个问题，本项目基于兼具加密与计算双重能力的同态加密体制，设计了高效的全同态加密方案，可根据实际密文计算次数设置合理的方案参数。基于所提方案，利用其Somewhat特性实现同态比较协议及同态求最大值协议，再将其应用于实现密文超平面决策分类、朴素贝叶斯分类及决策树分类。为研究医疗诊断中疾病的诱发因素，基于多密钥全同态加密方案，本项目设计了两种外包型安全多方k-means聚类方案。此外，本项目将同态加密的应用范围由移动医疗网络扩展到车联网，调研总结了车联网中基于同态加密的安全基本运算、数据聚合、数据查询和其它计算类型，并讨论研究了未来可能面临的挑战和问题。本课题研究移动医疗网络的隐私保护密文计算，有助于推进移动医疗网络的安全发展，丰富同态加密技术的研究与应用，具有重要的理论意义与实践价值。