From Gradient Clipping to Structural Refinement: Improving DPSGD for Medical Image Segmentation
For practitioners of differentially private medical image segmentation, this work identifies limitations of existing DPSGD methods and offers a practical improvement, though the gains are incremental.
The paper evaluates recent DPSGD variants (Auto-S, NSGD, PSAC) for medical image segmentation, finding that prior assumptions about gradient alignment do not consistently hold, and proposes an adaptive DP-Morph variant that improves segmentation quality under differential privacy constraints.
Medical image segmentation is widely used for disease detection but relies on sensitive data, raising privacy concerns as trained models can leak information. Differential privacy, typically implemented via Differential Private Stochastic Gradient Descent (DPSGD), provides a solution, though at the cost of reduced utility. Recent DPSGD variants, including Automatic clipping (Auto-S), Normalised SGD with perturbation (NSGD), and Per-sample adaptive clipping (PSAC), have shown promise in image classification, but their behavior in medical segmentation remains underexplored. We evaluate these methods across binary and multi-class tasks and analyze gradient alignment, showing that prior assumptions, particularly for PSAC, do not consistently hold. We further demonstrate that combining clipping strategies with morphological refinement improves segmentation quality under privacy constraints. Finally, we propose an adaptive DP-Morph variant that captures class-specific structures and enhances performance in multi-class settings.