Federated Explainable CNN–Vision Transformer Framework for Privacy-Preserving Multi-Center Lung Cancer Diagnosis Using CT Imaging

Lung cancer remains the leading cause of cancer-related mortality worldwide, with survival rates strongly dependent on early detection. Computed tomography (CT) imaging plays a critical role in identifying pulmonary nodules; however, manual interpretation is time-consuming and subject to inter-observer variability. Deep learning–based diagnostic systems have demonstrated strong performance under controlled conditions, yet their clinical deployment is constrained by data privacy regulations, poor cross-institution generalization, and limited interpretability. This paper proposes a Federated Explainable CNN–Vision Transformer (FedX-CNN-ViT) framework for privacy-preserving, multi-center lung cancer diagnosis using CT imaging. The proposed system enables collaborative learning across healthcare institutions without sharing raw patient data, while integrating domain-robust representation learning and embedded explainability mechanisms. By combining federated optimization, hybrid CNN–Vision Transformer feature extraction, domain adaptation, and explanation consistency regularization, the framework simultaneously addresses privacy, robustness, and clinical transparency. Experimental evaluation on multi-source public CT datasets demonstrates improved diagnostic accuracy, enhanced cross-domain generalization, and clinically meaningful explainability compared to centralized and non-explainable baselines. The results indicate that the proposed framework provides a scalable and trustworthy foundation for real-world deployment of AI-assisted lung cancer screening systems.