Atlas 2: Pathology Foundation Models for Clinical Deployment

Atlas 2, Atlas 2-B, and Atlas 2-S are three newly created pathology vision foundation models aimed at addressing prior challenges related to performance, robustness, and computational efficiency that have limited their use in clinical settings. These models were trained using the largest dataset for pathology foundation models to date, which includes 5.5 million histopathology whole slide images sourced from Charité - Universitätsmedizin Berlin, LMU Munich, and Mayo Clinic. A thorough assessment across eighty public benchmarks revealed that these models achieved cutting-edge predictive performance and robustness while ensuring efficient resource use. This development helps connect research progress with real-world applications in computational pathology.