Dual-UQ is the proposed federated optimisation strategy that explicitly models both aleatoric and epistemic uncertainty during aggregation.

Aleatoric uncertainty captures noise and ambiguity inherent in the microscopy data itself, while epistemic uncertainty reflects model uncertainty arising from limited or non-representative local datasets.

By combining both forms of uncertainty, Dual-UQ aims to produce a more reliable global model under strong site heterogeneity. In the simulation dashboard, Dual-UQ appears as the highlighted dark-red curve in the ROC and precision–recall plots whenever DenseNet121 is selected.