The rapid adoption of advanced technology in healthcare has led to unprecedented sensitive patient data, raising serious issues about privacy, security and compliance with regulations such as HIPAA and GDPR. Traditional centralized machine learning methods depend on consolidating all raw data in one location, making healthcare systems susceptible to data leaks, unauthorized access and security breaches. Federated learning (FL) has emerged as a promising approach for healthcare institutions to collaborate in training models while keeping raw data private. FL enhances privacy protection, assists businesses in adhering to data protection laws and minimizes security risks by consolidating data in one location and sharing only model changes. This research critically evaluates the function of FL in safeguarding privacy and augmenting security within e-Healthcare systems. Important FL variants such as FedAvg and FedProx are examined to address issues arising when healthcare data is not IID across hospitals. Advanced methods such as differential privacy, safe aggregation, and trust-aware procedures are examined to enhance the resistance of FL against attacks and ensure a reliable model. A comparative analysis of centralized and decentralized FL designs presents the trade-offs in scalability, resilience and security. Lastly, the report highlights areas where additional research is necessary. It proposes future directions, emphasizing the integration of trust management, blockchain-based auditability and customizable FL models suitable for various patient types. The study shows that FL is a scalable, secure, and privacy-preserving framework for speeding up the use of AI in healthcare. Encouragement of collaborative innovation occurs alongside the protection of patient privacy.