Practical and Secure SVM Classification for Cloud-Based Remote Clinical Decision Services

Abstract

Support vector machine (SVM) classification techniques have been widely adopted for building clinical decision models. In cloud-based remote clinical decision services, a healthcare center outsources the clinical decision model to a cloud server, which then provides remote clinical decision services to end users. In this article, we propose a practical and secure SVM classification scheme ( <inline-formula><tex-math notation="LaTeX">${\sf SSVMC}$</tex-math></inline-formula> ) for cloud-based remote clinical decision services. Specifically, we first extract SVM decision rules from an SVM classifier. Then, we leverage symmetric key encryption to protect the confidentiality of medical data and prevent the cloud service provider from misusing intellectual property of the outsourced clinical model. Finally, we build encrypted indexes to achieve efficient SVM classification. We define a leakage function, formulate a security definition, and provide a simulation-based security proof for <inline-formula><tex-math notation="LaTeX">${\sf SSVMC}$</tex-math></inline-formula> . The performance analysis demonstrates that <inline-formula><tex-math notation="LaTeX">${\sf SSVMC}$</tex-math></inline-formula> achieves linear computational complexity when an SVM classifier (a.k.a., the clinical decision model) is pre-trained. The simulations evaluate the impact of several parameters on time costs. The experimental evaluations show the performance differences between <inline-formula><tex-math notation="LaTeX">${\sf SSVMC}$</tex-math></inline-formula> and several existing schemes in terms of time costs, storage costs, communication costs, and precisions in a real-world clinical dataset, which demonstrate that <inline-formula><tex-math notation="LaTeX">${\sf SSVMC}$</tex-math></inline-formula> is computationally efficient with high decision accuracy.