Artificial intelligence in radiology has been validated, regulated and deployed almost entirely on adult imaging. An AJR expert panel sets out what that means for paediatric practice, and the answer is more uncomfortable than a data-gap argument suggests.
Children are not small adults on imaging: anatomy and physiology change continuously through growth, so a model needs age-specific training data to be valid at any given age. That data barely exists — public paediatric datasets are scarce, institutional data is fragmented, the diseases are rare, reporting practice is heterogeneous, and external validation is largely absent. On top of that sit governance problems specific to children: consent for secondary use of a child's imaging, the off-label use of adult-trained models on paediatric studies, and the absence of post-deployment surveillance to catch drift. Reimbursement, the panel notes, is misaligned in a way that discourages anyone from fixing this.
The recommendation is an implementation roadmap and dedicated paediatric infrastructure and standards. For a department today the usable part is narrower and immediate: before a tool is switched on for paediatric studies, ask what age range it was validated in, on what data, and what happens when it is applied outside that range. In India, where paediatric imaging volume is high and formal validation infrastructure is thin, that question falls to the department rather than to a regulator.
- Ask every vendor for the age range and the paediatric validation data behind any tool used on children.
- Do not let an adult-validated model run silently on paediatric studies — gate it by age at the worklist.
- Set up local post-deployment monitoring; nobody else is doing surveillance for paediatric performance.
- Record consent arrangements for secondary use of paediatric imaging before contributing to any dataset.
- Treat absence of external validation as a reason to keep a tool advisory rather than autonomous.
Why it matters
Adult-trained models are already being applied to children's imaging by default, without anyone having decided that they should be.
Don't overread it
This is an expert panel narrative review setting priorities — it does not report paediatric error rates or quantify the harm it warns about.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for radiology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free