Overview

• Researchers analyzed nearly 2,000 digitized pathology slides from seven independent cohorts spanning Europe and the United States.
• The transformer-based model predicted multiple biomarkers directly from standard histology, matching or outperforming single-target approaches for markers such as BRAF, RNF43 and microsatellite instability (MSI).
• Results suggest shared morphological patterns reflect co-occurring mutations, with many alterations occurring more frequently in MSI tumors.
• The team says the approach could accelerate diagnostic workflows by helping clinicians prioritize patients for molecular testing and personalized treatment decisions.
• The international collaboration included TU Dresden’s EKFZ, University Hospital Augsburg, NCT Heidelberg, Fred Hutchinson Cancer Center, the Medical University of Vienna and Mayo Clinic, with plans to adapt the method to other cancers.