In healthcare, protecting patient privacy isn’t optional—it’s the law. Rules like HIPAA and GDPR make it hard for hospitals and clinics to share data, which gets in the way of building AI tools together. That’s a big deal, since machine learning models really need lots of varied data to work well across different groups of people. The legal boundaries are tight for good reason, but they certainly introduce significant challenges for implementing advanced analytics in this field. This research presents a federated learning (FL) framework that enables collaborative model training across multiple healthcare institutions without exposing sensitive patient data. To enhance privacy protections, the framework integrates Homomorphic Encryption (HE) to secure model updates during transmission and applies Differential Privacy (DP) to mitigate risks of gradient leakage and reconstruction attacks. The framework effectively supports decentralized learning scenarios with non-IID data distributions while maintaining strong model performance. We tested our approach on the Medical MNIST dataset, which contains X-ray images from various body regions. That matters because it shows we can protect sensitive patient data without sacrificing performance—making real-world clinical use much more achievable.