The FDA has granted breakthrough designation to two generative AI devices designed to interpret chest X-rays and automatically draft the radiology reports that radiologists typically write by hand. This represents a significant shift in how artificial intelligence is being deployed in medical imaging, moving beyond simple image analysis to full report generation. The two companies receiving this designation are Cognita, a Stanford researcher-founded startup acquired by Radiology Partners, and Aidoc, which announced its breakthrough status for a tool called First Read on June 25, 2026.

How Does Generative AI in Radiology Differ From Previous AI Tools?

For years, machine learning systems have analyzed medical images like X-rays and CT scans, but they worked in a limited way. Traditional AI would highlight a suspicious area or abnormality and flag it for a radiologist to review and interpret. The new generative AI tools represent a fundamentally different approach. Instead of just pointing out a finding, these large vision language models can process an entire chest X-ray image and draft many of the written findings that would normally require a radiologist's pen and keyboard.

This technological leap is powerful but also challenging. The FDA's traditional validation frameworks were built around simpler AI tools that made binary decisions or highlighted regions of interest. Generative AI that produces full narrative reports requires new ways of thinking about accuracy, safety, and clinical utility. The breakthrough designations signal that the FDA believes these tools have the potential to address an unmet medical need, but it also means the agency is navigating uncharted regulatory territory.

What Are the Practical Benefits for Radiologists and Patients?

The primary appeal of these tools is time savings. Radiologists spend significant portions of their day writing reports, a task that is cognitively demanding but also repetitive. If generative AI can draft accurate reports that radiologists then review and edit, it could free up time for more complex diagnostic work and reduce burnout. For patients, faster report generation could mean quicker access to their results and faster clinical decision-making by their physicians.

Aidoc's First Read tool has a specific focus: detecting and describing four life-threatening findings on chest X-rays. This narrow scope may make validation easier and the clinical benefit clearer. By concentrating on high-stakes conditions rather than trying to generate comprehensive reports for all possible findings, the tool can be more precisely evaluated and deployed in clinical workflows where it adds the most value.

Steps to Understand How These Tools Fit Into Clinical Practice

• Report Generation: The AI processes the entire chest X-ray image and generates a draft report with findings, descriptions, and clinical impressions that a radiologist reviews before finalizing.
• Radiologist Review: Unlike fully automated systems, these tools require a human radiologist to read, verify, and approve the AI-generated report before it is released to the patient's medical team.
• Regulatory Oversight: The FDA's breakthrough designation means the tools have shown promise in early testing, but they will still undergo rigorous clinical validation before widespread deployment in hospitals and imaging centers.
• Workflow Integration: Implementation will require changes to how radiology departments organize their work, including training radiologists on how to efficiently review and edit AI-generated reports.

What Challenges Remain for Validation and Deployment?

The FDA breakthrough designations are encouraging, but they also highlight a fundamental challenge: validating generative AI in clinical settings is harder than validating traditional diagnostic algorithms. With older AI tools, you could measure accuracy by comparing the tool's decision against a gold standard. With generative AI that produces narrative text, the question becomes more complex. Is a report "accurate" if it captures all the key findings but phrases them differently than a human radiologist would? How do you measure the quality of a generated report when there are many correct ways to describe the same clinical finding ?

These questions are not merely academic. They directly affect patient safety and clinical trust. If a radiologist misses an error in an AI-generated report because they are skimming it quickly, or if the AI generates plausible-sounding but incorrect findings, the consequences could be serious. The FDA and the companies developing these tools will need to establish clear validation protocols and ongoing monitoring systems to ensure that the time-saving benefits do not come at the cost of diagnostic accuracy.

The breakthrough designations for Cognita and Aidoc's tools represent an important moment in clinical AI. These are not the first AI tools in radiology, but they are among the first to move beyond detection and into full report generation. As these tools move toward clinical deployment, they will serve as a test case for how the FDA, radiologists, and health systems can safely integrate generative AI into high-stakes medical workflows.