Privacy-Preserving Similarity Retrieval over Medical Images

With the swift progress in medical imaging technologies, content-based image retrieval (CBIR) in the healthcare sector has become essential for assisting in disease diagnosis, drawing significant interest from both academia and industry. However, the complex nature of medical images, which often contain sensitive patient data, underscores the need for privacy-preserving CBIR as a critical issue. Although numerous privacy-preserving CBIR frameworks have been proposed, they mainly focus on countering known-background attacks (KBA) and do not sufficiently protect image privacy in outsourced environments. To address this gap, we introduce a novel privacy-preserving Mahalanobis distance comparison (PMDC) method that aims to enhance the precision of medical image retrieval. This method is subsequently combined with the Mahalanobis distance-based fuzzy C-means (FCM-M) algorithm to develop TAMMIE, a scheme that provides precise and privacy-preserving medical image retrieval on encrypted data. TAMMIE allows image owners to securely upload their images and indexes to cloud servers while enabling users to query these servers without compromising the confidentiality of their searches. Extensive security analyses demonstrate that our schemes are resilient against attacks beyond KBA. Furthermore, comprehensive empirical evaluations on two real-world datasets and one synthetic dataset validate the efficiency of TAMMIE.