Authors: Dr. Nidhi Mishra, Sunil Vishwakarma, Sahil, Sneha Pandey, Shirish Shukla
Abstract: The growing integration of artificial intelligence (AI) in healthcare has greatly enhanced clinical decision-making and predictive capabilities. However, conventional centralized training approaches introduce significant concerns related to data privacy, security, and regulatory compliance. Patient data, often distributed across multiple healthcare institutions, cannot be easily shared due to strict privacy laws and ethical considerations. To overcome these limitations, this study presents a secure and scalable federated AI framework designed for privacy-preserving healthcare analytics, allowing collaborative model development without the need for centralized data collection. The proposed system employs federated learning to build a global model by combining locally trained updates from decentralized healthcare nodes, ensuring that sensitive patient information remains within institutional boundaries. To strengthen security and reliability, the framework incorporates secure aggregation techniques, encryption-based protection of model updates, and anomaly detection methods to defend against adversarial threats and data poisoning attacks. Additionally, the architecture supports scalability through adaptive client selection and communication-efficient update mechanisms, making it well-suited for large-scale and heterogeneous healthcare environments. Experimental results using distributed healthcare datasets indicate that the proposed federated AI approach achieves performance comparable to traditional centralized models while substantially minimizing privacy risks and communication costs. These findings demonstrate the potential of the framework to enable secure, compliant, and efficient analytics across distributed medical systems. Overall, this work establishes a practical pathway for deploying trustworthy AI solutions in real-world healthcare settings while safeguarding patient confidentiality.
DOI: http://doi.org/10.5281/zenodo.20679368
