Multimodal Federated AI for Trustworthy and Personalized Healthcare

The increasing availability of heterogeneous healthcare data has accelerated the development of artificial intelligence (AI) systems for clinical decision support and personalized care. However, centralized multimodal AI remains constrained by data silos, privacy regulations, and limited generalizability, raising concerns about trust, fairness, and accountability. Federated learning offers a promising decentralized alternative, yet its integration with multimodal AI introduces additional technical, organizational, and governance challenges that remain insufficiently synthesized. This review provides a state-of-the-art analysis of multimodal federated AI for trustworthy and personalized healthcare. We examine advances in multimodal learning and federated systems, analyze their intersection in clinical settings, and highlight how multimodality amplifies challenges related to data heterogeneity, privacy, explainability, and robustness. Distinct from algorithm-centric surveys, we frame trustworthiness as a system-level property encompassing privacy, fairness, robustness, explainability, and governance. We introduce a unifying framework that organizes design choices across data, model, and system layers, enabling systematic evaluation of clinical trade-offs beyond predictive accuracy. Finally, we identify persistent evaluation gaps and outline future directions, including longitudinal personalization and federated foundation models. By centering trust and clinical readiness, this review clarifies the design and deployment pathways for multimodal federated AI systems that are not only high-performing, but also equitable, transparent, and clinically actionable.