Waiv released two robustified digital pathology foundation models, Phaet (based on Phikon-v2) and Mascaret (based on Midnight-12k), on Hugging Face for research use. The models were fine-tuned using a model-agnostic technique designed to remove their fragility to variations in slide scanners, laboratory procedures, and tissue staining.
Our readBiopharma R&D teams and healthcare researchers can leverage these open-source foundation models to build computational pathology tools that generalize safely across multi-institutional datasets without silently losing accuracy due to site-specific technical artifacts.