HOPPR Foundation Model Library

These models reside within the HOPPR® AI Foundry

Disclaimer: Developers are responsible for making any necessary modifications, validating model performance in the final product, and obtaining any applicable regulatory marketing authorizations before commercialization. HOPPR provides tools and component-level documentation to support regulatory preparation and alignment.

ViT Foundation Model

HOPPR® MC Chest Radiography

  • Modality: Chest X-ray (frontal PA/AP views)
  • Anatomy: Chest – Lungs, heart, pleura, mediastinum, ribs, chest wall
  • Model Type: Vision Transformer (ViT) Foundation Model
  • Task: Classification fine-tuning with model score output. Internal validation on 24 findings, range 0.77–0.99 AUC.
  • Availability: Fine-tuning and inference via the HOPPR® AI Foundry and/or Forward Deployed Services

Read our press release on the MC Chest Radiography Model.

MC Chest Radiography

Foundation Model

Supports binary classification fine-tuning with model score output

~12.2M Images
~6.1M Studies
24 Findings
Internal validation · ROC-AUC
Median range
0.91
range 0.77–0.99 across 24 findings
Narrative VLM

HOPPR® MC Chest Radiography

  • Modality: Chest X-ray (frontal PA/AP, lateral)
  • Anatomy: Chest – Lungs, heart, pleura, mediastinum, ribs, chest wall, and devices (pacemaker, tubes)
  • Model Type: VLM – Vit encoder + Q-Former + modernBERTdecoder
  • Description: Generates descriptive, structured textual language outputs derived from chest radiography representations for research, development, and evaluation workflows.
  • Availability: Gain model access through HOPPR Forward Deployed Services and our Foundry API to run inference and test outputs against your data. Work with FDS to make targeted modifications scoped to your use cases and requirements.

Read our blog on the performance and evaluation metrics of our MC Chest Radiography Narrative Model.

Read our press release on the MC Chest Radiography model launch.

ViT Foundation Model

HOPPR® EB 2D Mammography

  • Modality: 2D Mammography (FFDM and/or 2D synthetic | CC and MLO views)
  • Anatomy: Breast (bilateral; left and right)
  • Model Type: Vision Transformer (ViT) Foundation Model with LoRA adapters
  • Task: Classification (cancer, density, pacemaker). Internal validation: ROC-AUC 0.90 (cancer), 0.94 (density), 0.99 (pacemaker). Supports laterality labeling and includes 5,400 pathology-proven studies.
  • Availability: Fine-tuning and inference via the HOPPR® AI Foundry and/or Forward Deployed Services

Read our press release on the EB 2D Mammography Foundation Model release.

EB 2D Mammography

Foundation Model

Supports binary classification including: cancer detection, density assessment, and pacemaker identification

~24M Images
~4M Studies
Internal validation · ROC-AUC
Cancer detection
0.90
Density
0.94
Pacemaker ID
0.99
Narrative VLM

HOPPR® EB 2D Mammography

  • Modality: 2D digital mammography (FFDM and synthetic 2D from DBT)
  • Anatomy: Breast (bilateral; left and right laterality)
  • Model Type: 2B-parameter Vision-Language Model (VLM)
  • Description: Generates descriptive, structured textual language outputs derived from 2D mammography representations for research, development, and evaluation workflows.
  • Availability: Gain model access through HOPPR Forward Deployed Services and our Foundry API to run inference and test outputs against your data. Work with FDS to make targeted modifications scoped to your use cases and requirements.

Read our press release on the EB 2D Mammography Model.

Narrative VLM

HOPPR® EF Chest CT

  • Modality: Chest CT (3D volumetric CT)
  • Anatomy: Lung nodule characterization and aortic measurement, covering the range of findings in a routine chest CT study
  • Model Type: 3D Chest CT with a three-component VLM architecture trained end-to-end on chest CT–report pairs: a frozen V-JEPA 2.1 ViT-L vision encoder adapted with LoRA, a fully trainable 6-layer Q-Former bridge with 64 learnable query tokens, and a ModernBERT causal decoder.
  • Description:  Processes full 3D CT volumes through a six-component pipeline combining vision-language generation, anatomical segmentation, and pulmonary nodule detection into a single structured output. Trained on a large proprietary dataset of chest CT studies from multiple U.S. clinical sites, with deliberate coverage of rare but serious conditions such as aortic injury, pulmonary embolism, and pneumothorax.   Descriptive imaging language is generated from chest CT, including lung nodule characterization and aortic measurement, covering the range of findings in a routine chest CT study. 
  • Availability: Gain model access through HOPPR Forward Deployed Services and our Foundry API to run inference and test outputs against your data. Work with FDS to make targeted modifications scoped to your use cases and requirements.

Read the press release on the EF Chest CT Narrative Model

3rd Party Models Available for Inference via the Foundry

Upon Request (Fine-tune with FDS)

CheXagent
2-3b srrg impression: Multimodal VLM

  • Modality: Chest X-ray (frontal AP/PA Lateral Optional)
  • Anatomy: Chest – lungs, heart, pleura, mediastinum, ribs, chest walls
  • Availability: Available for inference via AI Foundry

Findings Card Impressions Card

MedGemma 4B (Google)
Vision-language model

  • Modality: Chest X-ray, CT, MRI, histopathology, fundus, dermatology
  • Anatomy: Multi-organ, general medical imaging
  • Availability: Available for inference via AI Foundry
Documentation

MedImageInsights (Microsoft)
Vision-language embedding model

  • Modality: X-ray, CT, MRI, Mammo, ultrasound, OCT, histopathology, fundus, dermatology
  • Anatomy: multi-organ, general medical imaging
  • Availability: Available for inference via AI Foundry
Model Card

NVIDIA NV Reason 
CXR 3B: parameter VLM

  • Modality: Chest X-ray
  • Anatomy: Chest – lungs, heart pleura, mediastinum, ribs, chest wall
  • Availability: Available for inference via AI Foundry
Model Card

NVIDIA NV Generate
CT: 3D latent diffusion model

  • Modality: CT : full body up to 127 anatomical classes
  • Anatomy: Full body (up to 127 anatomical classes)
  • Availability: Available for inference via AI Foundry
Model Card

NVIDIA NV Generate
MR: 3D latent diffusion model

  • Modality: MRI T1, T2, FLAIR, SWI
  • Anatomy: Brain, abdomen, cardiovascular, and respiratory
  • Availability: Available for inference via AI Foundry
Model Card

RadFM
Multimodal VLM

  • Modality: 2D/3D radiology: X-ray, CT, MRI, and others
  • Anatomy: Brain, head/neck, thorax, spine, abdomen, pelvis, upper and lower limbs
  • Availability: Available for inference via AI Foundry
Project Page

 

Third Party Models Third Party Models

Disclaimer: Developers are responsible for making any necessary modifications, validating model performance in the final product, and obtaining any applicable regulatory marketing authorizations before commercialization. HOPPR provides tools and component-level documentation to support regulatory preparation and alignment.

Documentation on third-party models is available on their respective websites.

Streamlining responsible medical imaging AI

Accelerate Your Clinical AI Development

The HOPPR® AI Foundry strips away technical complexity, empowering your team to build compliant medical imaging solutions with speed and confidence.