Adult-Only AI Biopsy Model Accurately Diagnoses Paediatric Coeliac Disease — Research Library

Key Highlights

An AI model trained only on adult duodenal biopsies diagnosed coeliac disease in children with near-perfect accuracy, including slides from hospitals and scanners it had never seen.
The validation study, published in eClinicalMedicine, addresses whether adult-derived medical AI transfers safely to paediatric patients.
Biopsy-based diagnosis remains variable, with reported Cohen's kappa values as low as 0.35 and between 0.419 and 0.888 across pathology settings.

Context and Core Dynamics

Coeliac disease is a gluten-triggered autoimmune disorder affecting roughly 1.4 percent worldwide on serological testing, with about 0.7 percent confirmed by biopsy; most cases go undiagnosed. Symptoms include chronic diarrhoea, weight loss, anaemia, fatigue, skin rashes and infertility; untreated disease carries significantly elevated lymphoma and adenocarcinoma risk. In children, anxiety, food-related hypervigilance, social restriction and impaired quality of life are increasingly recognised.

Biopsy-based histology remains the gold standard but is less reproducible than assumed. European Society guidelines acknowledge significant interobserver variability. In one digital study, the mean probability of two pathologists agreeing on a coeliac diagnosis from whole-slide images was 0.73, with Cohen's kappa 0.59 without clinical or laboratory data and only 0.80 with serological information. Earlier work reported kappa as low as 0.35 under Marsh-Oberhuber classification and between 0.419 and 0.888 across pathology practice settings.

Earlier multicentre work by the same team reached expert pathologist-level accuracy in adult duodenal coeliac biopsies, but paediatric tissue remained unexplored. One earlier paediatric attempt was confounded: environmental enteropathy images came from hospitals in Pakistan and Zambia, while coeliac and normal images came from a different US institution with different imaging methods, so pathology versus image provenance was unclear. Some researchers suggest paediatric and adult disease differ biologically in gene expression, clinical presentation and mucosal pathology. The new validation reports near-perfect paediatric accuracy even on slides from unseen hospitals and scanners.

Caregiving Application and Social Impact

Coeliac disease requires lifelong gluten-free management; diagnosis reshapes household routines, food purchasing and social participation. Family caregivers absorb that load alongside the child's anxiety and food-related hypervigilance. A reproducible, automated reading of duodenal biopsies could shorten diagnostic delay and give dietitians and paediatric teams firmer planning. Demand is likely to rise: Italy has become the first country to mandate nationwide paediatric screening, and referral pathways for children outside serology-based criteria lead to duodenal biopsy sampling of the bulb (D1) and descending duodenum (D2).

Ethics and Compliance Note

Deployment raises consequential questions. Paediatric histopathology data are sensitive, cross-border image sharing requires clear lawful bases under data protection rules, and models trained on adult cohorts require local revalidation before clinical reliance. Pathologist oversight, explainability and equitable access across resource-differing health systems are prerequisites, not refinements. Regulatory frameworks for clinical decision support for minors should demand prospective, multi-site evidence before adoption.