{"id":"arxiv:2501.05409","kind":"preprint","title":"Atlas: A Novel Pathology Foundation Model by Mayo Clinic, Charité, and Aignostics","authors":["Maximilian Alber","Stephan Tietz","Jonas Dippel","Timo Milbich","Timothée Lesort","Panos Korfiatis","Moritz Krügener","Beatriz Perez Cancer","Neelay Shah","Alexander Möllers","Philipp Seegerer","Alexandra Carpen-Amarie","Kai Standvoss","Gabriel Dernbach","Edwin de Jong","Simon Schallenberg","Andreas Kunft","Helmut Hoffer von Ankershoffen","Gavin Schaeferle","Patrick Duffy","Matt Redlon","Philipp Jurmeister","David Horst","Lukas Ruff","Klaus-Robert Müller"],"author_count":27,"year":2025,"published":"2025-01-09","venue":null,"cited_by":null,"doi":null,"arxiv_id":"2501.05409","nct_id":null,"source_url":"https://arxiv.org/abs/2501.05409","archive_url":"https://space-kitty.com/papers/arxiv:2501.05409","updated":"2025-01-10","pmid":null,"pmcid":null,"openalex_id":null,"pdf_url":"https://arxiv.org/pdf/2501.05409","abstract":"Recent advances in digital pathology have demonstrated the effectiveness of foundation models across diverse applications. In this report, we present Atlas, a novel vision foundation model based on the RudolfV approach. Our model was trained on a dataset comprising 1.2 million histopathology whole slide images, collected from two medical institutions: Mayo Clinic and Charité - Universtätsmedizin Berlin. Comprehensive evaluations show that Atlas achieves state-of-the-art performance across twenty-one public benchmark datasets, even though it is neither the largest model by parameter count nor by training dataset size.","abstract_license":"CC0-1.0 (arXiv descriptive metadata)","license":null,"is_open_access":false,"retracted":false,"peer_reviewed":false,"categories":["cs.CV","cs.AI","cs.LG"],"details":null,"provenance":"arxiv","attribution":"Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.","topics":[],"cited_in_dossiers":[]}