Adaptive Differential Privacy for Federated Medical Image Segmentation Across Diverse Modalities
Summary: arXiv:2604.06518v1 Announce Type: cross
Abstract
Large volumes of medical data remain underutilized because centralizing distributed data is often infeasible due to strict privacy regulations and institutional constraints. In addition, models trained in centralized settings frequently fail to generalize across clinical sites because of heterogeneity in imaging protocols and continuously evolving data distributions arising from differences in scanners, acquisition parameters, and patient populations. Federated learning offers a promising solution by enabling collaborative model training without sharing raw data.
However, incorporating differential privacy into federated learning, while essential for privacy guarantees, often leads to degraded accuracy, unstable convergence, and reduced generalization. In this work, we propose an adaptive differentially private federated learning (ADP-FL) framework for medical image segmentation that dynamically adjusts privacy mechanisms to better balance the privacy-utility trade-off.
Key Contributions
- Stabilizes training processes while improving Dice scores and segmentation boundary quality.
- Maintains rigorous privacy guarantees throughout the model training.
- Evaluates the effectiveness of ADP-FL across diverse imaging modalities and segmentation tasks.
Evaluation and Results
The proposed approach was evaluated across various segmentation tasks, including:
- Skin lesion segmentation in dermoscopic images.
- Kidney tumor segmentation in 3D CT scans.
- Brain tumor segmentation in multi-parametric MRI.
Compared with conventional federated learning and standard differentially private federated learning, ADP-FL consistently achieves:
- Higher accuracy in segmentation results.
- Improved boundary delineation of anatomical structures.
- Faster convergence during training cycles.
- Greater training stability across diverse datasets.
Conclusion
The results indicate that ADP-FL approaches the performance of non-private federated learning under the same privacy budgets, demonstrating the practical viability of this framework for high-performance, privacy-preserving medical image segmentation. This advancement could lead to greater utilization of medical data while ensuring compliance with privacy regulations, ultimately enhancing patient care through improved diagnostic tools.
As federated learning continues to evolve, the integration of adaptive differential privacy stands out as a promising pathway for the future of medical imaging and other sensitive data applications.