Development of Federated Learning-Based AI Framework for Privacy-Preserving Medical Diagnostics in Cottage Hospital and Federal Polytechnic Ukana Clinic Akwa Ibom State

Abstract

This study developed and evaluated a federated learning-based artificial intelligence framework for privacy-preserving medical imaging diagnostics in two low-resource healthcare facilities in Akwa Ibom State, Nigeria. The objective was to improve diagnostic accuracy, operational efficiency, and patient data protection without centralizing sensitive medical information. A total of 3,395 chest X-ray and ultrasound images were collected and used to train lightweight convolutional neural networks under a federated learning protocol employing encrypted model aggregation and differential privacy mechanisms. Performance was benchmarked against manual diagnosis and centralized deep learning models. The federated global model achieved 91.6% diagnostic accuracy, representing a statistically significant improvement over baseline manual diagnosis (73.8%, p < 0.001). Diagnostic time was reduced by 75%, and energy consumption decreased by 37.5%. Privacy leakage simulations demonstrated substantial protection under ε-differential privacy constraints. Robustness testing confirmed stable performance under low-bandwidth conditions. Economic evaluation indicated a favorable return on investment within the first operational year. The findings demonstrate that federated AI frameworks can deliver clinically meaningful improvements while maintaining regulatory compliance and data sovereignty in resource-constrained healthcare environments. The study provides a scalable roadmap for secure AI-enabled diagnostics in developing regions.