Design and Evaluation of a Privacy-Enhanced Federated AI System for Healthcare Applications

Predictive analytics, personalized treatment strategies, and automated disease diagnosis have rapidly advanced due to the increasing integration of Artificial Intelligence (AI) in healthcare systems; however, the development of accurate AI models requires access to large-scale clinical datasets that are highly sensitive and governed by strict privacy regulations. Conventional centralized machine learning approaches aggregate patient data into a single repository, increasing the risk of data breaches and regulatory violations. To address this limitation, this study proposes a Federated Learning (FL) framework that enables multiple healthcare institutions to collaboratively train a deep neural network without sharing raw patient data. The primary objective is to design a distributed and secure learning architecture that ensures high predictive performance while preserving patient confidentiality. A federated deep neural network model was implemented for disease prediction using decentralized clinical datasets distributed across participating institutions, and secure aggregation protocols along with differential privacy mechanisms were incorporated to mitigate inference attacks. Experimental evaluation demonstrates that the federated model achieves comparable predictive accuracy to centralized training while significantly enhancing data security and reducing privacy risks, even under heterogeneous data distributions across institutions. The findings indicate that federated learning provides a scalable, trustworthy, and regulation-compliant solution for multi-institutional healthcare AI applications, contributing to secure collaborative medical intelligence and addressing critical ethical and legal challenges in healthcare data management.