Lessons Learned from Implementation of Multi Institutional Collaborative Platform for Lung Disease Screening by Means of Federated Learning in Indonesia

Abstract

Purpose: This study aims to reduce lung disease mortality by developing an automatic screening system that analyzes X-ray images and utilizes distributed image data storage, creating a collaborative platform among Indonesian hospitals while ensuring patient data privacy. Methods: Using Federated Learning, a decentralized machine learning approach, hospitals build local models with their data, which are aggregated into a global model on a central server without compromising confidentiality. An innovative system for archiving medical images is also introduced, which anonymizes, secures, and curates data for training marchine learning based diagnosis systems. Results: The Federated Learning implementation resulted in a privacy preserved detection model that aggregates models from multiple hospitals while keeping patient data secure and remains in their silos. The archiving system successfully stores anonymized medical images and created a valuable dataset for CAD development. Conclusion: This work advances machine learning in healthcare, prioritizing patient privacy while enhancing X-ray analysis and collaborative model development. By addressing technical and ethical challenges, this framework sets a new standard for responsible AI in healthcare, with potential for application in other imaging modalities and diseases, aiming to revolutionize medical diagnostics.