Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Atlas 2 -- Foundation models for clinical deployment

Maximilian Alber, Timo Milbich, Alexandra Carpen-Amarie, Stephan Tietz, Jonas Dippel, Lukas Muttenthaler, Beatriz Perez Cancer, Alessandro Benetti, Panos Korfiatis, Elias Eulig, Jérôme Lüscher, Jiasen Wu, Sayed Abid Hashimi, Gabriel Dernbach, Simon Schallenberg, Neelay Shah, Moritz Krügener, Aniruddh Jammoria, Jake Matras, Patrick Duffy, Matt Redlon, Philipp Jurmeister, David Horst, Lukas Ruff, Klaus-Robert Müller, Frederick Klauschen, Andrew Norgan

TL;DR

Atlas 2 advances pathology AI by delivering a 2B-parameter tile-based Vision Transformer trained on 5.5 million histopathology WSIs from three institutions, with distilled variants Atlas 2-B and Atlas 2-S that dramatically reduce compute and inference time. The work comprehensively evaluates on eighty public benchmarks spanning morphology, molecular, and survival prediction, demonstrating state-of-the-art performance, robustness, and efficiency across tasks and datasets. Robustness is explicitly quantified with PathoROB, Plismbench, and Patho-Bench, and distillation achieves substantial gains in practicality for clinical deployment. Collectively, the Atlas 2 family offers scalable, robust, and resource-efficient foundation models that are well-positioned for clinical translation in digital pathology.

Abstract

Pathology foundation models substantially advanced the possibilities in computational pathology -- yet tradeoffs in terms of performance, robustness, and computational requirements remained, which limited their clinical deployment. In this report, we present Atlas 2, Atlas 2-B, and Atlas 2-S, three pathology vision foundation models which bridge these shortcomings by showing state-of-the-art performance in prediction performance, robustness, and resource efficiency in a comprehensive evaluation across eighty public benchmarks. Our models were trained on the largest pathology foundation model dataset to date comprising 5.5 million histopathology whole slide images, collected from three medical institutions Charité - Universtätsmedizin Berlin, LMU Munich, and Mayo Clinic.

Atlas 2 -- Foundation models for clinical deployment

TL;DR

Atlas 2 advances pathology AI by delivering a 2B-parameter tile-based Vision Transformer trained on 5.5 million histopathology WSIs from three institutions, with distilled variants Atlas 2-B and Atlas 2-S that dramatically reduce compute and inference time. The work comprehensively evaluates on eighty public benchmarks spanning morphology, molecular, and survival prediction, demonstrating state-of-the-art performance, robustness, and efficiency across tasks and datasets. Robustness is explicitly quantified with PathoROB, Plismbench, and Patho-Bench, and distillation achieves substantial gains in practicality for clinical deployment. Collectively, the Atlas 2 family offers scalable, robust, and resource-efficient foundation models that are well-positioned for clinical translation in digital pathology.

Abstract

Pathology foundation models substantially advanced the possibilities in computational pathology -- yet tradeoffs in terms of performance, robustness, and computational requirements remained, which limited their clinical deployment. In this report, we present Atlas 2, Atlas 2-B, and Atlas 2-S, three pathology vision foundation models which bridge these shortcomings by showing state-of-the-art performance in prediction performance, robustness, and resource efficiency in a comprehensive evaluation across eighty public benchmarks. Our models were trained on the largest pathology foundation model dataset to date comprising 5.5 million histopathology whole slide images, collected from three medical institutions Charité - Universtätsmedizin Berlin, LMU Munich, and Mayo Clinic.
Paper Structure (39 sections, 2 figures, 7 tables)

This paper contains 39 sections, 2 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Pathology foundation models
  4. Robustness of pathology foundation models
  5. Efficient pathology foundation models for clinical deployment
  6. Data and Methods
  7. Dataset and Preprocessing
  8. Model Framework and Compute Environment
  9. Evaluation Protocols
  10. HEST
  11. eva
  12. PathoROB
  13. Plismbench
  14. Patho-Bench
  15. Additional benchmarks
  16. ...and 24 more sections

Figures (2)

  • Figure 1: (A) The results show that Atlas 2 is the best performing model. Additionally Atlas 2-B and 2-S have similar prediction performance as contenders but are up to a magnitude more efficient. The performance average is based on morphology prediction tasks from evakaiko.ai2024eva and molecular prediction tasks from HEST jaume2024hest in Table \ref{['tab:main_results']}. The dots show the processing speed of the respective model for a 224$\times$224 pixel image on an L4 GPU. The value for Pluto-4G pluto4g is approximated by taking the fastest speed of same sized models kaplan2025openmidnighthoptimus0xu_whole-slide_2024. (B) The results show that Atlas 2, Atlas 2-B, and Atlas 2-S are the most robust models and also the state of the art performance-robustness Pareto front. The plot shows the performance average from (A) and the robustness average from Table \ref{['tab:main_results']}. (C) The data shows that the Atlas 2 models are the performance-resource efficiency Pareto front. The plot shows the performance average of (A).
  • Figure 2: (A) - (E) show the average performance per evaluation framework kaiko.ai2024eva, jaume2024hest, pathobench, koemen2025pathorob, and plismbench. Atlas 2 is the leading contender on all frameworks.