Efficient Multi-disease Privacy-Preserving Medical Pre-Diagnosis Based on Partial Homomorphic Encryption

With the development of the Internet, there are more and more sensitive information on medical data, and direct use will result in the leakage of relevant information. These privacy issues largely limit the development of the medical industry, and online medical diagnosis services can break the time and region restrictions. In response to the existing privacy requirements, we use the random forest of machine learning to train the classifier. Compared with other classification models, the random forest classifier has higher accuracy and can process large-scale medical data. In the process of interaction between medical service providers and medical users, SHE (symmetric homomorphic encryption) method and Boneh-Lynn-Shacham(BLS) short signature algorithm are used to ensure the privacy and non-tampering of data during the interaction. Since both the random forest and the user query vector is in the state of ciphertext, we design a security comparison algorithm to ensure that the comparison can be completed without revealing privacy. Futhermore, a disease risk list can be obtained, which can achieve multi-disease diagnosis. We also prove that the proposed protocol is secure and efficient by security analysis and efficiency analysis.