The increasing digitization of healthcare systems and the proliferation of medical IoT devices have heightened the risk of cyber threats, making robust, transparent, and privacy-preserving security solutions essential. In this paper, we propose B-FXAI, a novel framework that integrates Federated Learning (FL), Blockchain, and Explainable Artificial Intelligence (XAI) for real-time and accountable threat detection in smart healthcare environments. By leveraging FL, sensitive patient data remains local while allowing collaborative training across institutions. A permissioned blockchain ensures trust, immutability, and verifiability of model updates, while SHAP-based explainability provides interpretable insights into model predictions. Our experiments on benchmark cybersecurity datasets TON_IoT, CIC-IDS2017, and MedBIoT demonstrate that B-FXAI achieves superior accuracy (up to 96.2%), reduces false positives (down to 2.7%), and maintains high interpretability coverage. The proposed framework outperforms centralized and baseline federated models, offering a scalable, transparent, and privacy-preserving solution for critical infrastructures such as healthcare systems.