Homomorphic encryption for secure healthcare artificial intelligence

Abstract

As artificial intelligence (AI) becomes deeply embedded in healthcare, safeguarding sensitive medical data against cyber threats is paramount. Homomorphic encryption (HE) has emerged as a promising solution for privacy-preserving computation, yet practical deployment remains hindered by trade-offs in accuracy, latency, and computational cost. This paper presents a comprehensive PRISMA-guided scoping review on the integration of HE into healthcare AI systems that utilize homomorphic encryption algorithms, with a focus on healthcare applications and edge-deployed architectures. By analyzing 31 applied studies across edge, cloud, and federated settings, HE schemes are classified according to their suitability for specific data types, system constraints and pipeline stages. In addition, key research questions to guide future work in secure healthcare artificial intelligence development are identified. Furthermore, a structured decision matrix is synthesized that aligns encryption strategies with healthcare-specific requirements, offering practitioners an actionable guide for secure AI deployment. This Systematization of Knowledge (SoK) advances the cybersecurity community’s understanding of HE’s role in protecting medical AI. Unlike prior surveys, this work goes beyond cataloguing schemes by systematizing their mapping to cybersecurity threat models and deployment contexts. The proposed decision matrix is intended as a design-support tool synthesizing reported empirical trends rather than a substitute for deployment-specific benchmarking.