ATHENA: An Accountable, Trustworthy Healthcare Ecosystem for Federated AI Nursing and Analytics

Authors: Mohan Siva Krishna Konakanchi

DOI: https://doi.org/10.37082/IJIRMPS.v8.i6.232816

Short DOI: https://doi.org/hbnx94

Country: United States

Full-text Research PDF File:   View   |   Download

Abstract: Federated Learning (FL) presents a powerful so- lution for training robust AI models on sensitive healthcare data distributed across multiple institutions, preserving patient privacy by design. However, the real-world application of FL in healthcare is challenged by significant statistical heterogeneity, disparities in data quality, and the potential for unreliable contributions from participating silos, which can compromise the integrity of the collaboratively trained model. To address these critical issues, we propose ATHENA (Accountable, Trustworthy Healthcare Ecosystem for Federated AI Nursing and Analytics), a novel framework designed to ensure accountability and ro- bustness in cross-silo healthcare collaborations. At the core of ATHENA is a multi-dimensional trust metric that dynamically assesses the reliability of each participant. This metric evaluates not only the performance of local model updates but also their consistency with the global learning objective and a novel clinical plausibility score, which serves as a privacy-preserving proxy for data quality. By leveraging this trust score, ATHENA’s aggre- gation mechanism adaptively weights contributions, effectively mitigating the impact of faulty or divergent silos. Furthermore, recognizing that clinical adoption hinges on both accuracy and physician trust, ATHENA incorporates a formal methodology for quantifying and optimizing the trade-off between a model’s predictive performance and its clinical explainability. We validate ATHENA through extensive simulations on a large-scale radio- logical dataset, demonstrating that it significantly outperforms standard federated learning approaches in realistic non-IID and noisy environments and provides a principled tool for navigating the crucial balance between performance and interpretability.

Keywords: Federated Learning, Trustworthy AI, Health- care AI, Explainable AI (XAI), Accountability, Medical Infor- mastics.

Paper Id: 232816

Published On: 2020-11-08

Published In: Volume 8, Issue 6, November-December 2020

Share this