Where radiology expertise meets AI innovation.
The HOPPR AI Foundry is a secure developer platform built specifically for medical imaging AI by practicing radiologists and AI engineers.
It brings together everything teams need to build, fine-tune, and validate imaging AI so they can focus on their use case, not on assembling the environment around it.
Request AccessPre-Trained on Real-World Data. Tailored to Your Workflow.
HOPPR’s Foundation Model Library gives development teams access to proprietary models built on real-world imaging data, alongside a curated selection of leading third-party models. Results are delivered in structured formats (standardized JSON) and text for easy integration.
Included in the Foundry are:
- Proprietary, large-scale Vision Transformer (ViT) foundation models, trained on tens of millions of labeled and de-identified medical images.
- Built on self-supervised learning, these models learn visual representations that generalize across diverse datasets, patient demographics, sites, and imaging systems, making them a meaningful starting point for fine-tuning.
- Each model can be fine-tuned to a specific workflow based on use case and patient population, precisely what point solutions can’t offer.
- Narrative Vision Language Models (VLMs) that generate narrative language describing imaging characteristics, giving downstream development teams a structured output to integrate into radiology workflow applications.
- Third-party models are also available for running inference on request.
Flexible Data Options
You can bring your own imaging data or use HOPPR’s curated, labeled, and validated datasets to fine-tune foundation models for specific use cases.
This flexibility supports experimentation, reproducibility, and faster iteration without compromising compliance.
HOPPR Labeled and Validated Datasets
HOPPR® AI Foundry provides one of the largest repositories of labeled and validated imaging data with verified provenance. These datasets are annotated, validated, and balanced with positive and negative examples for reliable training.
Foundation Model LibraryDataset Highlights:
Mammography
5400 pathology-proven studies, includes laterality labeling (e.g., left vs. right breast)
Chest Radiography
25 datasets featuring 200,500 unique studies for PA / AP projections
Granularity
Laterality and metadata provenance unique to HOPPR’s dataset design
Consistency
Enables you to reproduce HOPPR’s baseline model performance
Fine-tune, Evaluate, and Run Inference
HOPPR’s ViT foundation models are designed to be fine-tuned. Rather than deploying a fixed model across every environment, teams can adapt them to their specific use case, patient population, and imaging context, producing models that reflect where and how they’ll be used.
Within the Foundry, teams can select their data options, fine-tune models using HOPPR’s tools and compute, evaluate performance against their own specifications, manage model versions, and run inference on pre-trained models.
Intuitive User Interface
For teams that prefer a more guided experience, an intuitive UI is layered over HOPPR’s APIs, providing end-to-end access to fine-tuning, evaluation, and version management without requiring extensive DevOps support.
Teams that prefer to work directly via API can do so. For those that need additional clinical or technical support, Forward Deployed Services partners with you, meeting you where you are, and filling resource gaps as needed.
QMS-aligned framework within a secure, scalable infrastructure
The Foundry was developed under a quality management system (QMS) aligned with ISO 13485, IEC 62304, ISO/IEC 42001, and ISO 14971. This framework provides the quality, safety, and traceability foundation to support regulatory preparation and documentation generation throughout the development lifecycle.
The Foundry operates within a secure, HIPAA-compliant environment supporting fine-tuning and inference via API, usage-based billing, and end-to-end traceability.
HOPPR® provides a compliant foundation designed specifically for ingesting, fine-tuning, validating, and hosting imaging models — enabling faster, responsible AI development at scale.
Deploy models into production-ready environments
The HOPPR AI Foundry makes it simple to connect your fine-tuned or third-party models to your existing applications and workflows, ensuring your team can extend innovation securely and at scale. From our Foundry, you can easily integrate your fine-tuned or third-party models into your application using developer-friendly RESTful APIs.
Disclaimer: You are responsible for making any necessary modifications, validating model performance in the final product, and obtaining any applicable regulatory marketing authorizations before commercialization.
What our partners say
“For us, the HOPPR AI Foundry is transformative. We can leverage our data and clinical expertise to fine-tune models, rapidly test in production, and then scale as we see improved outcomes. Foundry is instrumental in the holy grail of adaptive learning in clinical workflow.”
Sham Sokka
Chief Operating and Technology Officer · DeepHealth
“The next generation of medical imaging AI will combine multimodal reasoning with the ability to generate high-fidelity clinical data. Platforms like the HOPPR AI Foundry enable developers to train and deploy medical imaging on NVIDIA accelerated computing with the performance and scale required for healthcare innovation.”
David Niewolny
Director of Business Development for Healthcare and Medical · NVIDIA
Accelerate Medical Imaging AI Development
Unite foundation models, curated data, and secure infrastructure in a single ecosystem. The HOPPR® AI Foundry strips away technical complexity, empowering your team to build compliant medical imaging solutions with speed and confidence.