HOPPR has released the HOPPR™ MC Chest Radiography Narrative Model, a vision language model designed to translate chest X-ray images into descriptive, structured text. The Chicago-based company is positioning the model as a foundational software component for developers building radiology reporting tools and image-based workflow applications, not as a finished product for direct clinical deployment.
HotSpot Take
HOPPR’s chest X-ray narrative model gives developers a validated, traceable language layer for radiology AI — a foundational component the field has lacked as vision-language models proliferate with uneven reliability.
Giving Radiology AI a Language Layer
AI-assisted radiology reporting connects medical image analysis with structured clinical language.
For more than a decade, AI in medical imaging has functioned primarily as a detection engine: flagging findings, highlighting anomalies, and surfacing patterns for a radiologist to interpret and describe. The reporting step (converting image findings into the structured language of a radiology report) has remained largely a manual workflow. HOPPR’s MC CXR Narrative Model targets that gap directly.
The model processes standard chest X-rays, including both frontal and lateral views, and generates descriptive, structured language output that developers can incorporate into reporting or workflow applications. According to the company, training data was drawn from a large set of chest X-ray reports and evaluated across a wide range of common patterns, with sampling designed to reflect the diversity of conditions seen in real-world clinical environments.
“We just gave medical images a voice,” said Roger Boodoo, MD, Medical Director of AI at HOPPR and a practicing radiologist. “For over a decade, AI gave us a second set of eyes but left us to do all the talking. By providing models that enable developers to build applications that translate images into natural language workflows, we’re reducing AI friction and giving radiologists their time back.”
Built for Developers, Not Direct Deployment
The MC CXR Narrative Model is not deployed as a standalone clinical application. It is designed for integration by development teams building their own radiology AI tools, with HOPPR providing deployment support through its Forward Deployed Services (FDS) unit. FDS works directly with partner organizations to evaluate and adapt the model to specific use cases, workflows, and data environments, with initial applications targeting workflow augmentation, training, and research.
The design reflects a deliberate infrastructure-first philosophy. Key capabilities include training data traceability (HOPPR maintains records of what data was used to build the model), along with version control that allows development teams to lock model versions for consistency and reproducibility across deployment stages. Organizations can also fine-tune the model against their own data to improve alignment with local reporting conventions and patient populations.
“The industry is in an arms race where a state-of-the-art model or point solution is outdated in weeks. What matters is flexible, underlying infrastructure that teams can adapt to their specific environment.” — Khan Siddiqui, MD, Co-Founder and CEO, HOPPR
“The industry is in an arms race where a state-of-the-art model or point solution is outdated in weeks,” said Khan Siddiqui, MD, co-founder and CEO of HOPPR. “What matters is flexible, underlying infrastructure that teams can adapt to their specific environment. This model is built with flexibility in mind: it’s a component that organizations can shape to their workflows and data with the traceability and validation behind it to support responsible deployment.”
Traceability as a Technical Priority
The emphasis on traceability and version control is not incidental. Vision language models applied to radiology have drawn increasing scrutiny over a tendency toward hallucination, specifically the generation of clinically plausible-sounding language that does not accurately reflect the image being analyzed. Researchers at the University at Buffalo recently described the problem in a study accepted at the 2026 Medical Imaging with Deep Learning Conference, finding that AI vision-language models tend to drift away from the image as they generate longer reports, relying on learned associations rather than visual evidence. Separately, Stanford researchers found that some frontier AI models could score well on medical benchmarks without ever analyzing actual images.
HOPPR’s approach, which emphasizes traceable training data, performance benchmarking against internal standards, and the capacity for developer-directed fine-tuning, is designed to support the validation process that responsible clinical deployment requires. The company has not published independent peer-reviewed benchmarks; performance claims are based on internal evaluation, according to the release.
Expanding the HOPPR AI Portfolio
The MC CXR Narrative Model extends the product line HOPPR introduced with the HOPPR™ AI Foundry, the company’s secure development platform for building, fine-tuning, validating, and hosting medical imaging AI. The Foundry debuted at the Radiological Society of North America’s 2025 Annual Meeting and is designed to provide developers with curated datasets, traceable development workflows, and regulatory-aligned infrastructure.
The narrative model represents a move from development tooling into deployable model assets: components that developer partners can pull into their own application stacks rather than build from scratch. HOPPR positions both the Foundry and the MC CXR Narrative Model within a broader mission to advance transparent, scalable AI for medical imaging.
Chest radiography occupies a particular place in that mission: it is one of the highest-volume imaging procedures in clinical practice, and automated report generation has emerged as a more active research area in medical imaging AI. As covered in HealthTech HotSpot’s medical imaging roundup from HIMSS26, the broader market continues to bifurcate between enterprise infrastructure platforms and clinical AI companies developing the algorithms and deployment frameworks that run on top of them. HOPPR’s narrative model sits at that intersection: a model asset designed for developer integration, backed by infrastructure built for traceability.
For health systems and radiology practices evaluating AI-assisted reporting tools, the path to accessing HOPPR’s model runs through its developer partners rather than direct procurement. How quickly those partnerships translate into commercially available reporting applications will determine the model’s near-term clinical footprint.
— This original article was created with AI support.
