pFM: Pathology Foundation Models

A portfolio of proprietary foundation models designed to capture how disease biology is represented in tissue.

Contact us

Transforming routine H&E tissue slides into deep biological insights

Pioneered by Paige and now driving innovation at Tempus, Pathology Foundation Models (pFM) use advanced vision AI to transform computational pathology. Trained on millions of real-world images, pFM uncovers the hidden biological signals within tissue morphology. This enables a wide range of applications, including predicting biomarkers, patient outcomes, and drug efficacy directly from standard H&E slides.

Models require appropriate validation and market authorization, as required by applicable law, prior to clinical deployment.

Our pathology foundation models

Virchow Family

Virchow is a family of pathology vision foundation models that create fine-grained representations of tissue morphology.

Model specifications

VIRCHOW

Description
The first million-slide foundation model for cancer, pioneering vision models in computational pathology

 

Training Dataset
1.5M H&E WSIs

 

Network Size
632 million parameters

 

Magnifications
20x magnifications

 

Licenses
Open Source
Apache 2.0 License

VIRCHOW2

Description
Expanded multi-scale training (H&E + IHC), boosting precision across 40+ tissue types

 

Training Dataset
3.1M H&E and IHC WSIs

 

Network Size
632 million parameters

 

Magnifications
5x-40x magnifications

 

Licenses
Open Source
CC-BY-NC-ND 4.0

VIRCHOW 2G

Description
Our largest model, optimized for high-stakes pathology AI

 

Training Dataset
3.1M H&E and IHC WSIs

 

Network Size
1.8 billion parameters

 

Magnifications
5x-40x magnifications

 

Licenses
Proprietary

VIRCHOW 2G-MINI

Description
A lightweight alternative, distilled from Virchow2G, for high-throughput applications

 

Training Dataset
3.1M H&E and IHC WSIs

 

Network Size
21.6 million parameters

 

Magnifications
5x-40x magnifications

 

Licenses
Proprietary

How it’s used:

  • Diagnostic tasks such as cancer detection, classification, and subtyping across both common and rare tissue types

  • Phenotype-genotype mapping enabling biomarker detection, expression and regulation

  • Novel biomarker discovery, including biomarkers that predict payload resistance and overall therapeutic response

Learn more about Virchow↗
Learn more about Virchow2↗

PRISM Family

PRISM is a family of multimodal vision and language foundation models that combine pathology images and clinical reports for enhanced whole-slide context and clinical reasoning.

Model specifications

PRISM

Description
A vision-language model for slide-level histopathology analysis and diagnostic report generation

 

Training Dataset
587k H&E WSIs
~195k associated diagnostic pathology reports

 

Network Size
~558M parameters

 

Magnifications
20x magnification

 

Licenses
Open Source
CC-BY-NC-ND-4.0

PRISM2

Description
An advanced vision-language model aligning tissue histomorphology with clinical dialogue and diagnostic reasoning

 

Training Dataset
2.3M H&E WSIs
685k associated diagnostic pathology reports

 

Network Size
~4B parameters

 

Magnifications
20x magnification

 

Licenses
Open Source
CC-BY-NC-ND-4.0

How it’s used:

  • Diagnostic tasks, including pan-tissue cancer detection and complex histological subtyping

  • Predict patient survival outcomes including progression-free, recurrence-free and disease-specific survival

  • Predict therapeutic benefit beyond traditional biomarkers to evaluate response to targeted treatments versus standard of care

  • Promptable via natural-language for capabilities such as diagnostic question-answering, report summarization, and report completion

Learn more about PRISM↗
Learn more about PRISM2↗

Case studies

Explore our case studies below to see how pFM performs with real-world data and scenarios.

Cancer Diagnostics
A person gazes at falling blue and orange digital data streams.

Identification of Invasive Lobular Carcinoma (ILC)

To overcome the low interobserver agreement and subjectivity associated with traditional histologic labels in invasive lobular carcinoma (ILC), a deep learning model was trained on whole-slide images using an objective genetic marker, CDH1 bi-allelic mutations, as the ground truth.

 

The model established a new paradigm for AI-based cancer classification, achieving 96% accuracy in identifying ILC. Furthermore, the model successfully expanded beyond structural mutations to detect alternative CDH1 inactivating mechanisms, including epigenetic silencing, enabling a more precise and comprehensive characterization of the disease phenotype.

pFM
Digital Pathology
Breast
CDH1

Partnership Framework

We offer flexible frameworks to align our AI capabilities with your needs

Partner with us at any stage, from proof of concept to commercialization, on a customized project tailored to your unique needs.

Discuss an AI partnership→

Deployed Applications

Explore the AI-enabled applications powered by our pathology foundation models today.

Explore our digital pathology solutions→

Genomic biomarker prediction using H&E alone

*For research use only

Paige Predict* is an AI application that predicts the status of ~1,600 biomarkers across >500 genes from a single H&E slide, enabling accelerated genomic insights across cancer types. Predictions can help researchers link morphological phenotypes with genotypes, prioritize use of tissue and pre-screen samples for biomarkers of interest at scale.

Explore Paige Predict’s biomarkers→

AI application for pan-cancer detection

*For research use only

Paige PanCancer Detect* is a groundbreaking application capable of detecting suspicious tissue across more than 40 cancer types to support research with pre-screening of all biopsy and resected tissue. Developed on Virchow, this application demonstrates the generalizable power of our foundation models.

Read our publication↗

Featured Resources

Hear from our experts

White square with a folded top-left corner on a dark background.Webinar
Abstract background with subtle wavy lines transitioning from grey to blue.

The Next Wave of ADC Development: From Precision Engineering to AI-Driven Insights

Discover how AI and multimodal data are solving ADC development’s toughest challenges—from predicting payload-specific resistance to unlocking insights from standard H&E slides.

Simple white graphic of a single sheet of paper or document, slightly angled.Article
QA_-From-proof-of-concept-to-clinical-deployment_-How-foundation-models-are-accelerating-drug-development

Q&A: From proof of concept to clinical deployment: How foundation models are accelerating drug development

Tempus leaders discuss the practical application of multimodal foundation models, from de-risking programs and uncovering novel biomarkers to deploying AI-driven insights at the point of care.

Simple white graphic of a single sheet of paper or document, slightly angled.Article
Headline: Paige Prostate's IVDR approval and FDA authorization as the first AI.

The landmark journey of Paige Prostate: IVDR approval and building on a legacy of regulatory excellence as the first FDA-authorized AI

The journey to integrate artificial intelligence into clinical practice is defined by milestones that build trust and establish new standards of care. In 2025, Tempus achieved such a milestone: Paige Prostate received EU IVDR certification.

White square with a folded top-left corner on a dark background.Webinar
Three diverse people in circular headshots on a dark background, with a play button.

Unlocking foundation models: Our experience from proof of concept to deployment at scale

An executive discussion on leveraging multi-modal foundation models to accelerate your drug development journey.

Simple white graphic of a single sheet of paper or document, slightly angled.Article
Tempus article title: Q&A: How Tempus leverages AI for biomarker screening at scale.

From image to insight: How Tempus is leveraging AI to advance biomarker screening at scale

By leveraging Paige Predict to screen for genomic and phenotypic biomarkers, clinicians and researchers have rapid access to valuable information to prioritize testing and pre-screen patients at scale.

Simple white graphic of a single sheet of paper or document, slightly angled.Article
Tempus article discusses AI models in pathology for solving cancer's tissue scarcity problem.

Solving cancer’s tissue scarcity problem: How Tempus uses AI models in pathology to transform tissue analysis

Tempus AI research demonstrates how AI can analyze digitized pathology images to predict tissue sample viability for genomic testing, a task that pathologists struggle to perform consistently.

News

Three stylized human silhouette icons in white and gray, representing a group of people.Press Release
Abstract digital display showing glowing blue and orange genetic code sequences.

Tempus Study Published in Nature Medicine Demonstrates Best-in-Class Performance of PRISM2 Across Diagnostic and Prognostic Applications

Tempus’ multimodal foundation model offers diagnostic-grade precision for cancer detection and unlocks broader applications predicting biomarker status and patient prognosis.

Three stylized human silhouette icons in white and gray, representing a group of people.Press Release
Blurred figures in white suits interact with glowing digital data in a futuristic lab.

Tempus Announces Strategic Collaboration with Daiichi Sankyo to Advance AI-Driven Biomarker Discovery and Clinical Differentiation Across an ADC Clinical Program

Through this collaboration, Daiichi Sankyo will leverage Tempus’ proprietary foundation models and AI expertise, including PRISM2, a state-of-the-art, multimodal foundation model that combines pathology images and clinical data to generate rich diagnostic and predictive insights.

Three stylized human silhouette icons in white and gray, representing a group of people.Press Release
Unable to determine image content.

Tempus Announces the Launch of Paige Predict

Tempus announced the launch of Paige Predict, a suite of cutting-edge digital pathology applications that analyze H&E whole slide images to help inform testing decisions.

Three stylized human silhouette icons in white and gray, representing a group of people.Press Release
A light gray and white blurred background with soft gradients.

Tempus Announces the Acquisition of Paige

Paige’s strong foundation in pathology and AI is rooted in its proprietary dataset of almost 7 million digitized pathology slides that are clinically annotated.

Research

Stack of white rectangular shapes or papers with a subtle shadow effect.Manuscript
Tempus-Vision_Digital-path_3

End-to-end multimodal pathology foundation model with clinical dialogue

PRISM2 publication in Nature Medicine (2026).

Stack of white rectangular shapes or papers with a subtle shadow effect.Manuscript
Digital code overlaying a blurred image of a person working at a computer.

A Foundation Model for Clinical-Grade Computational Pathology and Rare Cancers Detection

Virchow publication in Nature Medicine (2024).

Stack of white rectangular shapes or papers with a subtle shadow effect.Manuscript
Blurred data center with server racks and an overlay of digital alphanumeric characters.

Screen Them All: High-Throughput Pan-Cancer Genetic and Phenotypic Biomarker Screening from H&E Whole Slide Images

Preprint introducing Paige Predict, developed using Virchow2 (2025).

Stack of white rectangular shapes or papers with a subtle shadow effect.Manuscript
Blurred person wearing glasses, surrounded by a transparent blue digital data overlay.

Virchow2: Scaling Self-Supervised Mixed Magnification Models in Pathology

Preprint introducing Virchow2, Virchow 2G and Virchow2G Mini (2025).

Stack of white rectangular shapes or papers with a subtle shadow effect.Manuscript
Blurred abstract digital data or code with blue and light blue elements on a dark background.

PRISM: A Multi-Modal Generative Foundation Model for Slide-Level Histopathology

Preprint introducing PRISM (2024).

White graphic of a folded document or paper on a dark background.Abstract
Tempus-Vision_Digital-path_2

H&E-Based Deep Learning Model Predicts Immune Phenotypes in NSCLC

A deep learning model accurately identifies tumor-infiltrating lymphocytes (TILs) in non-small cell lung cancer from H&E slides, matching pathologist performance in distinguishing inflamed from non-inflamed immune phenotypes.

White graphic of a folded document or paper on a dark background.Abstract
Hands in a lab coat interacting with a glowing holographic display of biological cells.

Deep Learning Identifies FGFR Alterations from H&E Whole Slide Images in Bladder Cancer

A deep learning model trained on H&E slides predicts FGFR alterations in bladder cancer, potentially streamlining patient selection for targeted therapy NGS screening.

White graphic of a folded document or paper on a dark background.Abstract
Scientist in lab coat and black gloves working with advanced laboratory equipment.

Imaging-based histological features are predictive of MET alterations in Non-Small Cell Lung Cancer

MET is a proto-oncogene whose somatic activation in non-small cell lung cancer leads to increased cell growth and tumor progression

White graphic of a folded document or paper on a dark background.Abstract
Automated multi-channel pipette accurately dispenses red liquid into a laboratory plate.

Medical Image Computing and Computer Assisted Intervention – MICCAI 2021

Deep Orthogonal Fusion: Multimodal Prognostic Biomarker Discovery Integrating Radiology, Pathology, Genomic, and Clinical Data

This is AI-enabled precision medicine

This is the future of healthcare.