ADP-FL-MedSeg: Adaptive Differential Privacy for Federated Medical Segmentation

Researchers Puja Saha and Eranga Ukwatta have proposed a new framework called ADP-FL-MedSeg to address critical challenges in medical image analysis. Large volumes of medical data remain underutilized due to strict privacy regulations and the difficulty of centralizing distributed data across different clinical sites. While federated learning allows collaborative model training without sharing raw data, adding differential privacy often degrades accuracy and stability. This new approach introduces adaptive differential privacy within federated learning, dynamically adjusting privacy mechanisms to balance privacy guarantees with model utility. The framework was evaluated across diverse imaging modalities, including skin lesion segmentation in dermoscopic images, kidney tumor segmentation in 3D CT scans, and brain tumor segmentation in multi-parametric MRI. Results indicate that ADP-FL significantly improves Dice scores, segmentation boundary quality, and training stability compared to conventional methods. It achieves performance levels approaching non-private federated learning while maintaining rigorous privacy standards. This development demonstrates the practical viability of high-performance, privacy-preserving medical image segmentation in real-world federated settings, potentially enabling better generalization across heterogeneous clinical environments.