Patient-Centered and Practical Privacy to Support AI for Healthcare

Abstract

The increasing integration of artificial intelligence (AI) in healthcare holds great promise for enhancing patient care through predictive modeling and clinical decision support. However, privacy concerns emerge when deploying and sharing AI models, as adversaries can exploit vulnerabilities to infer sensitive patient information. Differential privacy (DP) has been the state-of-the-art approach to mitigate these risks, yet its adoption in healthcare remains limited due to complex privacy needs and the trade-off between privacy guarantees and model utility. This vision paper highlights the challenges and potential research directions of creating patient-centered privacy solutions that are practical, flexible, and transparent. They include improving patient awareness and control, developing privacy-enhanced training mechanisms that respect diverse patient preferences, and enabling post-training unlearning to adapt to evolving privacy requirements. While healthcare serves as a critical use case, the strategies discussed in this paper are applicable to other privacy-sensitive domains, aiming to advance the development of privacy-preserving AI systems for real-world applications across other data-sensitive domains.

Ähnliche Arbeiten