点击次数：

论文类型：期刊论文

发表时间：2021-03-05

发表刊物：COMPUTERS & SECURITY

卷号：99

ISSN号：0167-4048

关键字：Cloud storage; Content-based image searching; Homomorphic encryption; Large-scale medical images; Secure feature extracting; Electronic healthcare systems

摘要：The real-time sharing and retrieval of medical data, such as medical imaging data, via cloud systems can facilitate timely medical/disease diagnosis, for example during pandemics (e.g., COVID-19). While encryption can be used to ensure that patients' private and medi-cal information are not accessible by unauthorised individuals, it is challenging for cloud servers to search for and locate encrypted medical images (e.g. those relating to similar medical conditions). In this paper, we propose a novel and practical classification and retrieval method to search for and locate relevant cases over encrypted images. Specifically, we construct a privacy-preserving Convolutional Neural Network (CNN) framework that allows the classification and searching of secure, content-based, large-scale encrypted images (including large-size medical images) with homomorphic encryption. We analyze the security of our proposed method to ensure that no sensitive information from the encrypted images is leaked. Using four real-world datasets (i.e., chest X-Ray images, retinal OCT images, blood cell images, and Caltech101 image set), we evaluate and demonstrate the utility of our privacy-preserving method for searching images performed as well as CNN-based classification and searching of original images. This is an important step towards practical automated clinical diagnoses. (C) 2020 Elsevier Ltd. All rights reserved.