Differential privacy representation geometry for medical image analysis

Researchers introduce DP-RGMI, a framework that analyzes how differential privacy affects medical image analysis by decomposing performance degradation into encoder geometry and task-head utilization components. The study across 594,000 chest X-ray images reveals that differential privacy alters representation structure rather than uniformly collapsing features, providing insights for privacy model selection.

• →DP-RGMI framework decomposes differential privacy performance loss into encoder geometry and task-head utilization metrics.
• →Analysis of 594,000 chest X-ray images shows differential privacy creates utilization gaps even when linear separability is preserved.
• →Differential privacy alters representation anisotropy in non-monotonic, dataset-dependent ways rather than uniformly degrading features.
• →The framework enables reproducible diagnosis of privacy-induced failure modes in medical imaging applications.
• →Correlation analysis reveals robust associations between performance and utilization across datasets but with initialization-dependent variations.