Foundation Models Reshape Radiology AI Landscape: Industry Leaders Navigate Development and Regulation

The radiology sector is witnessing a significant shift as companies race to develop and implement foundation models, a new breed of artificial intelligence (AI) that promises greater accuracy and faster development in medical imaging. This emerging technology is poised to transform the way radiologists work, potentially addressing ongoing challenges such as increasing workloads and clinician shortages.

The Rise of Foundation Models in Radiology

Foundation models, which are trained on vast datasets and can provide multiple outputs, are gaining traction in the medical imaging field. The Food and Drug Administration (FDA) has already authorized over 950 AI-enabled devices in radiology as of May 2025, highlighting the rapid advancement of this technology.

Khan Siddiqui, CEO and co-founder of AI startup HOPPR, explains the potential impact: "A patient comes in, and has a headache. A headache could be hemorrhage, could be cancer, could be nothing, could be trauma." Traditional AI models often focus on single conditions, but foundation models aim to cover a broader spectrum of possibilities, aligning more closely with real-world clinical scenarios.

HOPPR has developed a foundation model trained on tens of millions of chest X-ray images, spanning hundreds of health conditions. The company is taking an innovative approach by offering its model to other medtech companies as a basis for developing their own fine-tuned models.

Diverse Approaches to AI Development and Implementation

While HOPPR focuses on providing a foundation for others to build upon, companies like Aidoc are taking a different route. Aidoc has created its own CARE foundation model, a vision-language model built using CT and X-ray images, along with supporting clinical information such as notes, labs, and vitals.

Elad Walach, CEO of Aidoc, reports that the company has already received FDA clearance for two derivative models based on this technology: a rib fracture triage tool and another to detect aortic dissection. Walach emphasizes the improved accuracy and comprehensiveness of their new model compared to previous AI tools.

Radiology Partners, a national practice encompassing more than 3,400 sites, is piloting two different models developed in-house. Their Mosaic Reporting tool, which uses large language models and voice recognition to structure radiology reports, has been rolled out to about 316 radiologists. The more complex Mosaic Drafting tool, which combines language and vision models to interpret images and pre-draft X-ray reports, is being tested by 57 radiologists under institutional review board approval.

Regulatory Landscape and Clinical Integration

As the technology advances, companies are grappling with the regulatory landscape for AI in healthcare. Walach notes, "The way the FDA works today, it's still disease by disease," highlighting the challenge of gaining approval for more comprehensive AI tools.

The integration of these advanced AI models into clinical practice also requires careful consideration. Nina Kottler, Associate Chief Medical Officer for Clinical AI at Radiology Partners, cautions: "A lot of people think of the technology, because it's so capable, that … someone could just buy it and plunk it in and maybe do a little training. It's actually far from the truth." She emphasizes the importance of training, change management, and continuous monitoring in successfully implementing AI tools.

As foundation models continue to evolve, they are likely to play an increasingly significant role in radiology and other medical fields. However, their successful integration will depend on navigating regulatory hurdles, ensuring clinical efficacy, and carefully managing the transition in healthcare settings.