Fig. 7: Workflow for preparing training data and performing whole slide virtual staining.

(1) PARS and stained WSIs are aligned using the Warpy registration framework. (2) Tissue borders and artifacts are annotated in QuPath, and the slides are tiled into artifact-free patches. (3) These patches are used to train the virtual staining model. (4) Virtual staining is applied to full PARS WSIs using overlapping patch-wise inference, followed by linear blending and tissue masking to produce the final virtual stained slide. The example shown depicts mouse H& E gastrointestinal tissue.