Balancing Privacy and Accuracy: Federated Learning with Differential Privacy for Medical Image Data

Federated Learning with differential privacy is increasingly employed across various domains to enhance data privacy. In the medical field, particularly with medical image data, it provides a crucial layer of protection for patient information. However, this approach often compromises detection performance. This study investigates the application of a federated setup combined with differential privacy to achieve a more favorable balance between accuracy and privacy. We utilized two datasets to evaluate the model's performance. On the Brain Tumor dataset, the model achieved an accuracy of 0.78, a precision of 0.70, and a recall of 0.60, with an epsilon of 5.6. For the OASIS dataset of Alzheimer disease, the model demonstrated a higher accuracy of 0.91, a precision of 0.93, and a recall of 0.91, with an epsilon of 3.825. These results indicate that our approach can effectively maintain privacy while ensuring substantial predictive performance.