HOPPR launches chest X-ray narrative AI model

HOPPR has introduced a vision-language model that converts chest X-rays into descriptive, structured text for radiology reporting and workflow applications.

The HOPPR MC Chest Radiography Narrative Model is designed as a foundational software component for developers building radiology workflow tools, rather than as a stand-alone clinical reporting product. The model processes standard frontal and lateral chest X-rays and generates narrative language that can be integrated into downstream applications.

The company said the model is intended to support development of reporting tools, workflow applications, and image-based AI systems that need a structured language layer. That makes the product different from a binary classifier that only flags the presence or absence of a finding.

HOPPR said the model was built with version control and training-data traceability, two features the company is positioning as important for responsible AI development. In practice, those controls are meant to help developers understand which model version is being used and how training data support the model’s behavior.

The model is deployed with support from HOPPR’s Forward Deployed Services team. That group works with partners to evaluate and adapt the model for specific use cases, workflows, and data environments, according to the company.

For radiology teams, the central issue is whether vision-language models can produce text that is useful, reliable, and clinically safe in real-world reporting workflows. Chest X-rays are high-volume exams, and even partial workflow support could matter if the generated text can be reviewed efficiently and does not introduce new safety risks.

“HOPPR’s MC CXR Narrative Model is not just a product launch, it’s a step toward a new generation of radiology AI,” said Khan Siddiqui, MD, founder and CEO of HOPPR.

The launch also reflects a broader shift in imaging AI. Earlier radiology AI tools often focused on detecting or triaging specific findings. Newer vision-language models are being developed to connect imaging inputs with structured text, reporting support, summarization, and workflow automation.

That shift brings new questions. A model that generates narrative text must be evaluated not only for whether it identifies findings, but also for whether it describes them accurately, avoids hallucinated content, handles uncertainty, and fits into radiologist review without creating misplaced trust.

HOPPR said the model is part of its wider medical imaging AI portfolio. The company previously released the Marie Curie Chest Radiography Model, which supports partners in fine-tuning and deploying binary classifiers for chest X-ray images using API access and structured outputs.

The MC CXR Narrative Model extends that portfolio from classification toward image-to-text generation. Its practical value will depend on validation, integration, monitoring, and how developers use it inside clinical products.

The company did not describe the launch as a direct replacement for radiologist reporting. The safer interpretation is that HOPPR is offering a developer-facing component that could support future radiology applications after integration, testing, and clinical governance.