Researches On Medical Diagnosis Scheme With Data Privacy Preservation

In the era of artificial intelligence,the rapid development of machine learning algorithms and cloud computing technologies has provided possibility for the application of classification algorithms in medical field.Some medical diagnosis schemes based on classification algorithms have been proposed.Support Vector Machine(SVM)is a generalized linear binary classifier using the maximum interval hyperplane as the decision boundary.It can also be extended to fit multi-classification and nonlinear problems.The diagnosis of the patient’s disease can be accomplished by inputting the unlabeled sample,that is,the patient’s medical data into the SVM classifier,that is,the diagnosis model,the classification result is the diagnosis result.However,the medical data of patients and the diagnosis model data of the diagnosis service provider are sensitive.Once the patient’s medical data is leaked,it will endanger the patient’s health.Correspondingly,the leak of the diagnostic model will cause the economic loss of the diagnostic service provider.How to protect the private information of patients and service providers at the same time is an urgent problem to be solved.In view of the security problems in the diagnosis process,this paper studies the privacy protection SVM classification problem in outsourced cloud environment,and designs a privacy-preserving linear multi-class SVM medical diagnosis scheme based on homomorphic encryption.In order to solve the problem that the negative numbers in the input data cannot be processed by the encryption algorithm,this paper designs a negative number encoding method to convert negative numbers to positive numbers to realize the expression of negative numbers in plaintext space and linear calculation.In response to the need to select the maximum decision function value in the multi-class SVM,a secure maximum value selecting algorithm on the ciphertext is designed.Considering that the integers on the ciphertext in the existing schemes are time-consuming and serious,this paper improves the existing secure integer comparison protocol,speeds up the comparison,and improves the efficiency of classification.Experiments on the real-world dermatology medical data sets show that the privacy-preserving linear multi-class SVM medical diagnosis scheme has a diagnostic accuracy rate of 97.3%,and has a better performance than existing schemes.The privacy-preserving linear multi-class SVM medical diagnosis scheme based on homomorphic encryption uses single server structure,and there is a problem that the failure of the single server will cause the entire diagnosis system to collapse.In order to solve this problem and expand the support for nonlinear data,this paper introduces homomorphic secret sharing technology,and designs a privacy-preserving nonlinear multi-server SVM medical diagnosis scheme.This paper uses the Shamir’s threshold scheme in which the secret information can be recovered with a part of participants,ensuring that the diagnostic system can operate normally even if a certain number of cloud servers are down.In this paper,the additive homomorphism and multiplicative homomorphism based of Shamir’s threshold scheme extend the support of the privacypreserving SVM for nonlinear data.Experiments on the real-world breast cancer medical data sets show that the privacy-preserving nonlinear multi-server SVM medical diagnosis scheme has a diagnosis accuracy rate of 95.6% and a good diagnosis efficiency.