H-Optimus-1

The leading foundation model for histology

Accelerate your research with our state-of-the-art foundation model, excelling on a large variety of tasks.

Multi-scale Foundation models

Bioptimus is building the word’s first multi-scale, multimodal foundation model for biology, uncovering and connecting the patterns that occur across different biological data modalities.

H-optimus-1 is the first piece of that multimodal puzzle, the leading foundation model for histology.

Foundation models learn the underlying patterns in vast datasets then represent them numerically as ‘embeddings’.

This can then be used to drive medical, biological and pharmaceutical research.

H-Optimus-1 model

H-optimus-1 is a cutting-edge tool for researchers to advance biomedical research, development, and diagnostics.

Histology images

H-optimus-1 leverages a vast and diverse dataset of histology images, encompassing millions of samples across various tissue types and pathological conditions.

Patch tokens

Each histology image is systematically partitioned into smaller patches, which are then transformed into token embeddings.

Transformer encoder

The extracted patch embeddings are processed by a transformer encoder, which models their complex spatial and structural relationships.

Embedding

The transformer encoder generates a compact embedding that encapsulates critical morphological and contextual features of the histology images.

Downstream tasks

This embedding serves as input for various specialized downstream tasks. For each downstream application category (e.g., response to treatment prediction, biomarker prediction, grading & prognosis prediction, tissue & cell segmentation), several dedicated AI downstream models, or heads, can be built. These specialized models leverage the embedding to perform targeted analyses.

In the example, a patient’s tissue image is analyzed by a specific downstream AI model tailored for outcome prediction, enabling an accurate assessment—83% risk of relapse. This approach allows each downstream model to be independently optimized for its predictive task.

H-optimus-1 is trained on 1 million+ whole-slide images (WSI) from 800,000+ patients, ensuring a diverse and robust dataset for various medical research applications.

What makes H-Optimus-1 different?

Training Data

H-optimus-1 uses a very large and diverse training dataset, compared to other models.

1 Million+

Whole Slide Images

2 Billion+

Tiles

800,000+

Patients

3

Scanner Types

50+

Organ tissues

4,000+

Clinical practices

Model Power

H-optimus-1 is one of the largest foundation models for pathology with over 1 billion parameters, allowing it to capture complex histological patterns and generate powerful embeddings.

Performance

H-optimus-1 has leading performance metrics compared to other publicly released models, making it a strong backbone for any use-case leveraging digital pathology.

H-optimus-1 reaches state-of-the-art performance on a variety of benchmarks and in downstream applications, such as the prediction of biomarkers of response to therapies (microsatellite instability in colorectal cancer and HER2 status in breast cancer).

Slide-level task average performance (AUROC)

HEST average performance (Pearson correlation)

How H-Optimus-1 helps your research

Biomarker Discovery & Patient Stratification

Identify novel predictive biomarkers for targeted therapies and precision medicine.

Enhance patient stratification for clinical trials, ensuring the right patients are selected based on histopathological and molecular features.

AI-Powered Drug Discovery & Preclinical Validation

Analyze drug-tissue interactions at scale, improving toxicology assessments and response predictions.

Predict drug efficacy across different cancer subtypes, reducing early-stage attrition rates.

Accelerating Clinical Trials & Regulatory Approvals

Automate histopathological grading & disease progression tracking to enhance trial endpoints.

Support AI-assisted clinical decision-making by integrating H-optimus-1 with real-world histology data.

Facilitate regulatory submissions with AI-powered standardization of pathology image assessments, ensuring compliance with FDA and EMA requirements.

Enhancing Digital Pathology & AI-Assisted Diagnosis

Leverage AI for automated annotation and analysis of whole-slide images (WSI) to reduce pathologist workload and improve consistency.

Enable real-time pathology analysis in research hospitals to support faster decision-making.

Why H-Optimus-1?

Unmatched performance

Delivers industry-leading results across diverse benchmarks and applications.

Tailored precision

Fully customizable to your proprietary datasets, optimizing performance for specific therapeutic areas and tasks.

Uncompromising privacy

Ensures complete data privacy—no data sharing or model training on your information—compliant with top-tier regulations like GDPR and the EU AI Act.

Transparent approach

Offers full visibility into data sources, model architecture, preprocessing, and internal evaluations, supporting regulatory-grade device development.

Flexible licencing

Provides versatile licensing solutions, from API-based access to seamless integration into research and R&D pipelines.

Scalable deployment

Supports deployment on-premises or in private cloud environments ensuring compliance with regulations like GDPR, HIPAA, and FDA 21 CFR Part 11.

Work with us

License the model through a platform provider or contact us directly if you’re interested to licence H-optimus-1.

If you have specific use cases you’d like to explore, or data you’d like to fine-tune the model with, please get in touch.