When AI Development Becomes Clinical Practice: Radiology's Regulatory Reckoning
The integration of artificial intelligence into radiology is creating an unexpected identity crisis for medical institutions. As AI tools become embedded in diagnostic workflows, radiology departments are finding that the traditional separation between technology development and clinical practice has largely evaporated.
This convergence presents significant challenges. When an algorithm is continuously updated based on patient outcomes, is it a research project or a medical device? Hospitals deploying in-house developed AI models may be making regulatory determinations that typically fall under FDA oversight, often without the infrastructure to handle such decisions appropriately.
The STAT investigation highlights how radiologists increasingly function as both clinicians and AI developers, training algorithms on their own patient data and iterating on models based on real-world performance. While this accelerates innovation, it also raises questions about validation standards, liability, and patient consent.
The core tension lies in accountability. Traditional medical device regulation assumes a clear distinction between manufacturers and users. But when clinicians modify algorithms based on local patient populations, the line between user and developer becomes impossibly blurry. Patients may not realize their diagnostic data is fueling ongoing AI research, nor that the AI assisting in their care is itself an experimental iteration.
Experts suggest this regulatory gray zone requires new frameworks that acknowledge how modern AI systems evolve continuously rather than existing as static, validated products. Until such frameworks emerge, radiology departments will continue navigating an uneasy middle ground where innovation and patient protection remain in tension.