Verifiable and Privacy-Preserving Online Diagnosis Based on Multiclass SVM and CKKS Leveled Homomorphic Encryption

The development of the Internet of Medical Things has made online diagnostic systems an attractive application. However, the opacity and insufficient supervision of cloud computing can result in problems such as data leakage and unreliable outcomes in diagnostic systems. In order to solve these problems, this paper proposes a verifiable privacy-preserving online diagnosis scheme (VPOD) based on multiclass SVM. This solution uses CKKS leveled homomorphic encryption to build a secure inner product computation protocol and a secure decision function computation protocol to achieve privacy protection for multiclass SVM model parameters, patient medical data and diagnosis results. In addition, this article designs a verification mechanism for multiclass SVM model predictions to ensure the correctness of the results. Performance evaluation shows that VPOD is more functional, has low computational overhead for patients and hospitals, and has an accuracy of 98.64% on real data sets.

I. Introduction

In recent years, with the development of the Internet of Medical Things (IoMT) [1], online diagnosis has become a widespread data service. Many online diagnostic systems based on different machine learning algorithms have been proposed [2] – [4]. The multiclass linear SVM model is a popular machine learning algorithm suitable for disease diagnosis in the IoMT environment [5]. However, hospitals have limited computing power, so they usually outsource complex calculations to cloud service providers.