FedTrust-XAI: Algorithmic Trust in Federated Health Communication Systems through Explainability and Privacy-Preserving Breast Cancer Diagnosis

Abstract

When using Artificial Intelligence (AI) in Clinical Decision Support Systems (CDSS), two important and related problems arise: ensuring that federated learn- ing over institutional data silos does not violate patient privacy, and gaining clinician trust in AI models’ diagnostic suggestions. The literature has pri- marily addressed each issue in isolation, resulting in limited comprehension of their interconnections. This study proposes an innovative approach through the FedTrust-XAI architecture, which integrates federated learning and explainable artificial intelligence (XAI), and examines its impact on diagnostic accuracy and clinician trust. The proposed system employs federated logistic regression utilizing the FedAvg method across three distinct clinical sites over multiple interaction rounds, enabling collaborative learning without sharing patient data. Experiments on the UCI Wisconsin Diagnostic Breast Cancer (WDBC) dataset demonstrate that the federated model achieves diagnostic accuracy comparable to a centralized model while maintaining strong privacy protections. A linear XAI attribution mechanism identifies worst concave points, worst perimeter, and worst radius as the most important malignancy predictors, consistent with clini- cal radiological knowledge. A factorial experiment with 120 simulated physicians shows that federated framing combined with XAI explanations yields the high- est clinician trust (Trust Index = 4.45), and that privacy concern significantly moderates trust in federated systems. The proposed architecture achieves 99.12% accuracy and 99.23% AUROC, closely matching the centralized baseline (99.60% AUROC). The results demonstrate that privacy-preserving federated learning, model explainability, and clinician trust can be jointly optimized in healthcare AI systems