基于同态加密的医疗数据密文异常检测方法

Homomorphic encryption-based ciphertext anomaly detection method for e-health records

为了避免电子健康数据(electronic health records,EHRs)在异常检测过程中泄漏患者信息和诊断结果等问题,针对医院与患者敏感数据的隐私保护,提出了一种基于CKKS全同态加密的EHRs异常检测隐私保护模型.将医院以及患者的EHRs由CKKS算法实现浮点数同态加密,设计一个基于密文比较算法的协议,通过可信密钥服务器与第三方数据中心之间的通信建立密文态孤立森林模型.并利用CKKS算法的SIMD技术,实现密文数据在孤立森林模型上的异常检测,最终返回密文结果.理论分析和实验结果表明,所提出的方案能够保证EHRs的隐私安全.并且在不同的EHRs数据集上验证,该模型优于传统的明文异常检测算法和同类型密文异常检测算法,且在密文态上能够保持和明文孤立森林算法相近的检测效率,有较好的异常检测效果.

To avoid the leakage of patient information and diagnosis results in electronic health data(EHRs)in the process of anomaly detection,a privacy protection model for EHRs’anomaly detection based on CKKS fully homomorphic encryption of sensitive data of hospitals and patients is proposed.The EHRs of hospitals and patients are encrypted using the CKKS algorithm to achieve floating-point number homomorphic encryption.Then,a protocol based on the ciphertext comparison algorithm is designed to establish a ciphertext state isolation forest model through the communication between the trusted key server and the third-party data center.Using the SIMD technology of the CKKS algorithm,the anomaly detection of the ciphertext data on the isolation forest model is realized,and the ciphertext result is finally returned.The theoretical analysis and experimental results show that the proposed scheme can ensure EHR privacy and security.It is verified on the dataset that this model is superior to the traditional plaintext anomaly detection algorithm and the same type of ciphertext anomaly detection algorithm.The model can maintain detection efficiency similar to the plaintext isolation forest algorithm in the ciphertext state and has a good anomaly detection effect.