FedERFT: Improving federated learning through feature-enriched regularization and post-aggregation fine-tuning | Synapse