Abstract: Federated learning (FL) faces significant challenges in medical image analysis due to data heterogeneity among clients, where balancing personalization and generalization is challenging.