Differential privacy in medical imaging applications

This bookchapter delves into the application of Differential Privacy (DP) in the realm of medical imaging. It begins by outlining the challenges inherent in leveraging large datasets for AI development in medicine while adhering to privacy and confidentiality constraints. The concept of DP is introduced as a robust mathematical framework designed to provide rigorous privacy guarantees. Then, the unique properties of DP are discussed, including its resilience to post-processing, compositional nature, and the quantifiable approach to privacy management through a privacy budget. The chapter further explores the integration of DP in deep learning, particularly in the training of neural networks using differentially private Stochastic Gradient Descent (DP-SGD), and addresses the challenges of DP, including the privacy-utility trade-off and fairness considerations. Selected applications of DP in medical image analysis are highlighted, demonstrating its practical efficacy and necessity in protecting sensitive patient data. The chapter concludes with a discussion on the evolving nature of DP, underscoring its potential for more tailored privacy guarantees and higher model utility in complex medical imaging applications.

Download paper here