Federated Learning: Envisioning a Pathway to More Accessible and Equitable Cleft Care

Cleft lip and palate represents the most common congenital craniofacial anomaly worldwide, with low- and middle-income countries (LMICs) bearing nearly 84% of the global disease burden. Despite the instrumental role of global nongovernmental organizations (NGOs) in expanding access to cleft care across LMICs, patient databases and healthcare infrastructure remain highly fragmented across centers and organizations. Recent advances in machine learning and deep learning hold considerable promise for improving cleft care through data-driven clinical decision-making. However, privacy regulations present significant barriers to the centralized aggregation of patient data required to train robust artificial intelligence models. Federated learning (FL) offers a potential solution by enabling multiple institutions to train a shared global model without transferring raw patient data to a central server. Through mechanisms such as federated averaging, differential privacy, and homomorphic encryption, FL substantially mitigates privacy risks while producing models that perform comparably to models trained on centralized pool data. Existing implementations, most notably the Federated Tumor Segmentation initiative spanning 71 institutions across 6 continents, demonstrate the feasibility of this approach in complex healthcare settings. This narrative review proposes a potential framework for applying FL to global cleft care, outlining how existing organizational infrastructure established by prominent cleft NGOs can provide a foundation for a federated network. Five key stakeholder categories: global cleft NGOs, academic medical centers, technology partners, funding entities, and local cleft centers have been identified as essential to realizing this vision. Numerous challenges and barriers have also been identified before realizing this endeavor.